Posts in category: Amazon AWS

Generative AI has transformed customer support, offering businesses the ability to respond faster, more accurately, and with greater personalization. AI agents, powered by large language models (LLMs), can analyze complex customer inquiries, access multiple data sources, and deliver relevant, detailed responses.

In this post, we guide you through integrating Amazon Bedrock Agents with enterprise data APIs to create more personalized and effective customer support experiences. Although the principles discussed are applicable across various industries, we use an automotive parts retailer as our primary example throughout this post.

By the end of this post, you’ll have a clear understanding of how to do the following:

• Use Amazon Bedrock Agents to create intelligent, context-aware customer support bots
• Integrate enterprise data sources, such as inventory management and catalog systems, with agents using AWS Lambda
• Build customized chat interfaces using the Amazon Bedrock Agents API
• Implement a solution that can instantly cross-reference product specifications with catalogs, check real-time inventory, and provide detailed information to the end-user

Solution overview

To illustrate the potential of this technology, consider an automotive parts retailer. In this industry, finding the right components can be challenging, because it often involves navigating extensive catalogs and complex compatibility requirements. An automotive retailer might use inventory management APIs to track stock levels and catalog APIs for vehicle compatibility and specifications. Access to car manuals and technical documentation helps the agent provide additional context for curated guidance, enhancing the quality of customer interactions.

The solution presented in this post takes approximately 15–30 minutes to deploy and consists of the following key components:

• Amazon OpenSearch Service Serverless maintains three indexes: the inventory index, the compatible parts index, and the owner manuals index. These indexes enable efficient searching and retrieval of part data and vehicle information, providing quick and accurate results.
• Amazon Bedrock Agents coordinates interactions between foundation models (FMs), knowledge bases, and user conversations. The agents also automatically call APIs to perform actions and access knowledge bases to provide additional information.
• Amazon Bedrock Knowledge Bases enables you to use Retrieval Augmented Generation (RAG), a technique that enhances responses from LLMs by incorporating information from a data store. By setting up a knowledge base with your data sources, your application can query it to provide answers, either through direct quotes from the sources or through naturally generated responses based on the query results.
• A web application serves as the frontend interface where users can initiate parts lookup requests.

Ingestion flow

The ingestion flow prepares and stores the necessary data for the AI agent to access. The following diagram illustrates how it works.

The workflow includes the following steps:

1. Documents (owner manuals) are uploaded to an Amazon Simple Storage Service (Amazon S3) bucket.
2. Amazon Bedrock Knowledge Bases ingests these documents:
   1. The knowledge base is configured to use the S3 bucket as a data source.
   2. The data source is synchronized and the knowledge base detects new, modified, or deleted documents in the S3 bucket and updates accordingly.
   3. The documents are chunked into smaller segments for more effective processing. This solution uses fixed-size chunking, where you can configure the desired chunk size by specifying the number of tokens per chunk and an overlap percentage.
3. Each chunk is embedded by using an embedding model such as Cohere Embed on Amazon Bedrock to create vector representations (embeddings) of the text.
4. The embeddings are stored in the Amazon OpenSearch Service owner manuals index. OpenSearch Service is used as the vector store for efficient similarity searching. The embeddings, along with metadata about the source documents, are indexed for quick retrieval.

User interaction flow

The following diagram illustrates the user interaction flow.

1. A user interacts with the Car Parts Agent through a web application interface. They can ask questions like “What wiper blades fit a 2021 Honda CR-V?” or ”Tell me about part number 76622-T0A-A01.”
2. The web application sends the user’s query to the Amazon Bedrock agent using the InvokeAgent API. The agent, using Anthropic’s Claude 3 Sonnet, interprets the user’s query and determines the best course of action through chain-of-thought (CoT) reasoning. At this stage, the agent employs guardrails to make sure it stays within its defined scope and capabilities. Through a runtime process that includes preprocessing and postprocessing steps, the agent categorizes the user’s input. This allows it to handle out-of-scope questions or potentially harmful inputs appropriately, without attempting to answer beyond its capabilities or knowledge base. The agent then analyzes the query to extract key information such as vehicle details, part numbers, or general automotive topics. If the query is within scope, the agent proceeds; if not, it provides a response indicating it can’t assist with that particular request.
3. For general inquiries, the agent consults its knowledge base in Amazon Bedrock, which includes information from various car manuals. This allows the agent to provide context and general information about car parts and systems.
4. For specific part inquiries, the agent consults the action groups available to the agent and invokes the correct action (API) to retrieve relevant information. This invocation happens when the agent determines that it needs to run a specific action based on the user input.
   1. The Lambda function runs the database query against the appropriate OpenSearch Service indexes, searching for exact matches or using fuzzy matching for partial information. It can access the inventory index for specific part details or the compatible parts index for compatibility information.
   2. The Lambda function processes the OpenSearch Service results and formats them for the Amazon Bedrock agent.
5. The Amazon Bedrock agent takes the formatted results and generates a human-readable response, combining database information with its general knowledge to provide comprehensive answers.

The following diagram illustrates the workflow of the agent.

This diagram illustrates the agent’s workflow from user query to response generation, integrating knowledge base and API data to provide comprehensive answers and handle follow-up questions.

Developer tools

The solution also uses the following developer tools:

• AWS Powertools for Lambda – This is a suite of utilities for Lambda functions that generates OpenAPI schemas from your Lambda function code. It provides annotations for business logic, descriptions, and parameter validations, automatically producing JSON-serialized OpenAPI schemas for use with Amazon Bedrock Agents.
• AWS Generative AI Constructs Library – This is an open source extension of the AWS Cloud Development Kit (AWS CDK) that offers multi-service, well-architected patterns for quickly defining generative AI solutions. It provides constructs to help developers build generative AI applications using pattern-based definitions for your infrastructure.

Prerequisites

You should have the following prerequisites:

• An AWS account with the appropriate AWS Identity and Access Management (IAM) permissions to create Amazon Bedrock agents and knowledge bases, Lambda functions, and IAM roles
• Access to the following models hosted on Amazon Bedrock:
  • Anthropic’s Claude 3 Sonnet (model ID: anthropic.claude-3-sonnet-20240229-v1:0)
  • Cohere Embed English v3 (model ID: cohere.embed-english-v3)
• A local development environment with the following:
  • The AWS Command Line Interface (AWS CLI) installed and configured with appropriate permissions.
  • Python 3.9 or later
  • Node.js v20.x or later
  • The AWS CDK CLI installed

Deploy the solution

The following steps outline the process to deploying the solution using the AWS CDK. The complete source code for this solution is available in the GitHub repository.

1. Open your terminal and run the following commands to clone the GitHub repository to your local machine:

   git clone https://github.com/aws-samples/bedrock-agent-carpart-lookup.git
   cd bedrock-agent-carpart-lookup

2. Create and activate a Python virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows, use .venvScriptsactivate
   

3. Install the required Python packages:

   pip install -r requirements.txt
   

4. Use the AWS CDK CLI to deploy the solution:

   cdk deploy

During deployment, you may be prompted to approve IAM role creations and security changes. Review and approve these if you’re comfortable with the permissions. After deployment, the AWS CDK CLI will output the web application URL. Make note of this URL (as shown in following screenshot) to access and test the agent.

After you deploy the solution, you can verify the created resources on the Amazon Bedrock console. On the Agents page, you’ll notice a new agent called car-parts-agent.

Effective agent instructions are crucial for optimizing the performance of AI-powered assistants. A well-structured set of instructions should encompass several key components:

• Agent role – Define the assistant’s purpose, such as serving as a Car Parts Assistant that helps users find compatible parts and automotive information
• Agent actions – Outline primary tasks, such as identifying parts based on vehicle details, verifying compatibility, and providing technical specifications
• Agent guidelines – Establish rules for interaction, prioritizing accuracy and safety, clearly stating uncertainties, and using actions for searches
• Agent guardrails – Implement limits to make sure the agent operates safely and effectively, using relevant automotive knowledge to enhance user support

For example, the agent we deployed has been preconfigured with the following instruction:

The agent has two main components:

• Action group – An action group named CarpartsApi is created, and the actions it can perform are defined using an OpenAPI schema. Optionally, you can use Powertools for AWS Lambda to simplify the process of generating the OpenAPI schema. For more information, refer to the PowerTools documentation on Amazon Bedrock Agents. The OpenAPI schema used by this agent can be viewed on the following GitHub repo. The action group is then associated with a Lambda function containing the business logic for these actions.
• Knowledge base – This repository enhances the agent’s responses using RAG in Amazon Bedrock. It contains information from car manuals and technical documentation. When associating a knowledge base with an agent, you can optionally provide a description on how the agent can use the knowledge base. For this demo, we use the following description for the knowledge base:

  This knowledge base contains manuals and technical documentation about various car makes from manufacturers such as Honda, Tesla, Ford, Subaru, Kia, Toyota etc.

  □ Instructions

The agent employs CoT reasoning to process user queries, analyzing input against its instructions and evaluating actions based the OpenAPI provided and knowledge base description. When required information is missing, as determined by the OpenAPI schema’s specifications, the agent formulates questions to elicit necessary data from the user. This analysis and information gathering leads to a logical sequence of steps, including API calls and knowledge base queries. The resulting observation enhances the prompt for the FM, which then determines and runs the most effective actions.

For this post, we use the AWS CDK and the AWS Generative AI Constructs Library to create the Amazon Bedrock agent. This approach enables version-controlled, reproducible infrastructure as code (IaC). Alternatively, you can create agents using the AWS CLI or AWS Management Console. For more details, see create and configure agent manually.

Test and validate the solution

After you deploy the solution, you can test the agent either on the Amazon Bedrock console or through the application URL noted earlier. In the application, you can ask questions like “What wiper blades fit a 2021 Honda CR-V?” or “Tell me about part number 76622-T0A-A01.”

To illustrate the Amazon Bedrock agent’s reasoning capabilities, let’s examine the agent’s response to a user question “What wiper blades fit a 2021 Honda CR-V?” The workflow would consist of the following steps:

1. The agent analyzes the query, identifying key elements: Honda (make), CR-V (model), 2021 (year), and wiper blades (part category).
2. It then reasons about which API to use, selecting the get_compatible_parts function as the most appropriate for vehicle-specific part queries.
3. The agent maps the query elements to the function’s parameters, demonstrating its understanding of how to translate natural language to API calls.
4. The agent calls the carPartsApi/get_compatible_parts action, which queries the OpenSearch Service index for compatible parts. When an action group using a Lambda function is invoked, Amazon Bedrock sends a Lambda input event using a general format.
5. Upon receiving a response, the agent interprets the data, identifying relevant wiper blade options. The agent formulates a natural language response, recommending specific wiper blades. Similarly, Amazon Bedrock expects a response from your Lambda function to match a pre-defined format.

We get the following natural language response:

Based on the compatible parts search for a 2021 Honda CR-V in the “Wipers” category, here are the recommended wiper blade options:

• Wiper Blade Set (Part #76622-TLA-A11) – This is a genuine Honda wiper blade set that includes the driver and passenger side wiper blades for the CR-V.
• Rear Wiper Blade (Part #76630-TLA-A01) – This is the rear wiper blade specifically for the 2021 CR-V.
• You can also consider individual wiper blades like the Driver Side Wiper Blade (Part #76622-T0A-A01) or the Passenger Side Wiper Arm (Part #76630-T0A-A01) if you only need to replace one side.

This process showcases how Amazon Bedrock agents can intelligently integrate enterprise data APIs with AI-driven reasoning to provide accurate, context-aware responses to customer queries.

Clean up

To avoid future costs and clean up resources after you’re done exploring the solution, delete the resources you created by running the following command from your terminal (from the project directory):

cdk destroy

Key considerations

When implementing Amazon Bedrock Agents, consider the following factors to facilitate optimal performance and scalability:

• Agent design – Follow these recommendations when designing your agent:
  • Keep instructions focused and clear, with specific responsibilities for the agent
  • For complex use cases, consider multiple specialized agents rather than overloading a single one
  • Explore different FMs to find the best fit for your needs, considering both behavior and cost
• Action management – Consider the following recommendations for action management:
  • Define actions carefully, including only those that the agent should reliably perform
  • Use clear, descriptive names for actions to help the agent determine their relevance
  • Avoid overlapping actions to prevent confusion and conflicts during operation
• Testing – Make sure your testing includes the following steps:
  • Establish clear testing protocols
  • Identify common use case inputs and set accuracy targets
  • Define edge case inputs and agree on acceptable accuracy levels
  • Determine out-of-domain inputs where the agent should not respond
  • Automate tests and run them with system changes to verify consistency and reliability
• Performance optimization – Consider the following performance optimizations:
  • Break down complex operations into smaller actions to enhance response time and error handling
  • Implement a “fail fast” principle for invalid queries, allowing more time for complex tasks
• Security and compliance – Use Amazon Bedrock Guardrails to prevent the agent from generating harmful content or making unauthorized actions
• Cost management – Monitor usage-based pricing for token processing and storage, facilitating efficient resource allocation and cost management

Conclusion

Integrating enterprise data APIs with Amazon Bedrock Agents offers a powerful solution for streamlining customer support, as demonstrated in the automotive parts industry. This AI-driven approach enables rapid, accurate responses to complex queries, seamlessly integrates multiple data sources, and reduces staff workload while enhancing customer experience through context-aware interactions.

The solution discussed in this post can elevate customer support across various industries. By using Amazon Bedrock agents, organizations can create more efficient, accurate, and satisfying support experiences tailored to their specific needs. To explore how AI agents can transform your own support operations, refer to Automate tasks in your application using conversational agents.

About the Authors

Deepak Kovvuri is a Senior Solutions Architect supporting Automotive and Manufacturing Customers at AWS in the US Northeast. He has over 6 years of experience in helping customers architecting a DevOps strategy for their cloud workloads. Deepak specializes in CI/CD, Systems Administration, Infrastructure as Code and Container Services. He holds an Masters in Computer Engineering from University of Illinois at Chicago.

Kingston Bosco is a Senior Solutions Architect for Global Strategic Partners at AWS. He designs and implements solutions that optimize DevOps workflows, automate cloud operations, and improve infrastructure management for customers. He holds a Master’s in Information Systems. In his free time, he enjoys hiking with his dogs and playing soccer.

Read More

This post is co-written with Steven Craig from Hearst. 

To maintain their competitive edge, organizations are constantly seeking ways to accelerate cloud adoption, streamline processes, and drive innovation. However, Cloud Center of Excellence (CCoE) teams often can be perceived as bottlenecks to organizational transformation due to limited resources and overwhelming demand for their support.

In this post, we share how Hearst, one of the nation’s largest global, diversified information, services, and media companies, overcame these challenges by creating a self-service generative AI conversational assistant for business units seeking guidance from their CCoE. With Amazon Q Business, Hearst’s CCoE team built a solution to scale cloud best practices by providing employees across multiple business units self-service access to a centralized collection of documents and information. This freed up the CCoE to focus their time on high-value tasks by reducing repetitive requests from each business unit.

Readers will learn the key design decisions, benefits achieved, and lessons learned from Hearst’s innovative CCoE team. This solution can serve as a valuable reference for other organizations looking to scale their cloud governance and enable their CCoE teams to drive greater impact.

The challenge: Enabling self-service cloud governance at scale

Hearst undertook a comprehensive governance transformation for their Amazon Web Services (AWS) infrastructure. The CCoE implemented AWS Organizations across a substantial number of business units. These business units then used AWS best practice guidance from the CCoE by deploying landing zones with AWS Control Tower, managing resource configuration with AWS Config, and reporting the efficacy of controls with AWS Audit Manager. As individual business units sought guidance on adhering to the AWS recommended best practices, the CCoE created written directives and enablement materials to facilitate the scaled adoption across Hearst.

The existing CCoE model had several obstacles slowing adoption by business units:

• Extreme demand – The CCoE team was becoming a bottleneck, unable to keep up with the growing demand for their expertise and guidance. The team was stretched thin, and the traditional approach of relying on human experts to address every question was impeding the pace of cloud adoption for the organization.
• Limited scalability – As the volume of requests increased, the CCoE team couldn’t disseminate updated directives quickly enough. Manually reviewing each request across multiple business units wasn’t sustainable.
• Inconsistent governance – Without a standardized, self-service mechanism to access the CCoE teams’ expertise and disseminate guidance on new policies, compliance practices, or governance controls, it was difficult to maintain consistency based on the CCoE best practices across each business unit.

To address these challenges, Hearst’s CCoE team recognized the need to quickly create a scalable, self-service application that could empower the business units with more access to updated CCoE best practices and patterns to follow.

Overview of solution

To enable self-service cloud governance at scale, Hearst’s CCoE team decided to use the power of generative AI with Amazon Q Business to build a conversational assistant. The following diagram shows the solution architecture:

The key steps Hearst took to implement Amazon Q Business were:

1. Application deployment and authentication – First, the CCoE team deployed Amazon Q Business and integrated AWS IAM Identity Center with their existing identity provider (using Okta in this case) to seamlessly manage user access and permissions between their existing identity provider and Amazon Q Business.
2. Data source curation and authorization – The CCoE team created several Amazon Simple Storage Service (Amazon S3) buckets to store their curated content, including cloud governance best practices, patterns, and guidance. They set up a general bucket for all users and specific buckets tailored to each business unit’s needs. User authorization for documents within the individual S3 buckets were controlled through access control lists (ACLs). You add access control information to a document in an Amazon S3 data source using a metadata file associated with the document. This made sure end users would only receive responses from documents they were authorized to view. With the Amazon Q Business S3 connector, the CCoE team was able to sync and index their data in just a few clicks.
3. User access management – With the data source and access controls in place, the CCoE team then set up user access on a business unit by business unit basis, considering various security, compliance, and custom requirements. As a result, the CCoE could deliver a personalized experience to each business unit.
4. User interface development – To provide a user-friendly experience, Hearst built a custom web interface so employees could interact with the Amazon Q Business assistant through a familiar and intuitive interface. This encouraged widespread adoption and self-service among the business units.
5. Rollout and continuous improvement – Finally, the CCoE team shared the web experience with the various business units, empowering employees to access the guidance and best practices they needed through natural language interactions. Going forward, the team enriched the knowledge base (S3 buckets) and implemented a feedback loop to facilitate continuous improvement of the solution.

For Hearst’s CCoE team, Amazon Q Business was the quickest way to use generative AI on AWS, with minimal risk and less upfront technical complexity.

• Speed to value was an important advantage because it allowed the CCoE to get these powerful generative AI capabilities into the hands of employees as quickly as possible, unlocking new levels of scalability, efficiency, and innovation for cloud governance consistency across the organization.
• This strategic decision to use a managed service at the application layer, such as Amazon Q Business, enabled the CCoE to deliver tangible value for the business units in a matter of weeks. By opting for the expedited path to using generative AI on AWS, Hearst was never bogged down in the technical complexities of developing and managing their own generative AI application.

The results: Decreased support requests and increased cloud governance consistency

By using Amazon Q Business, Hearst’s CCoE team achieved remarkable results in empowering self-service cloud governance across the organization. The initial impact was immediate—within the first month, the CCoE team saw a 70% reduction in the volume of requests for guidance and support from the various business units. This freed up the team to focus on higher-value initiatives instead of getting bogged down in repetitive, routine requests. The following month, the number of requests for CCoE support dropped by 76%, demonstrating the power of a self-service assistant with Amazon Q Business. The benefits went beyond just reduced request volume. The CCoE team also saw a significant improvement in the consistency and quality of cloud governance practices across Hearst, enhancing the organization’s overall cloud security, compliance posture, and cloud adoption.

Conclusion

Cloud governance is a critical set of rules, processes, and reports that guide organizations to follow best practices across their IT estate. For Hearst, the CCoE team sets the tone and cloud governance standards that each business unit follows. The implementation of Amazon Q Business allowed Hearst’s CCoE team to scale the governance and security that support business units depend on through a generative AI assistant. By disseminating best practices and guidance across the organization, the CCoE team freed up resources to focus on strategic initiatives, while employees gained access to a self-service application, reducing the burden on the central team. If your CCoE team is looking to scale its impact and enable your workforce, consider using the power of conversational AI through services like Amazon Q Business, which can position your team as a strategic enabler of cloud transformation.

Listen to Steven Craig share how Hearst leveraged Amazon Q Business to scale the Cloud Center of Excellence

Reading References:

• Getting started with Amazon Q
• The Business Value of AWS Cloud Governance Services IDC white paper.

About the Authors

Steven Craig is a Sr. Director, Cloud Center of Excellence. He oversees Cloud Economics, Cloud Enablement, and Cloud Governance for all Hearst-owned companies. Previously, as VP Product Strategy and Ops at Innova Solutions, he was instrumental in migrating applications to public cloud platforms and creating IT Operations Managed Service offerings. His leadership and technical solutions were key in achieving sequential AWS Managed Services Provider certifications. Steven has been AWS Professionally certified for over 8 years.

Oleg Chugaev is a Principal Solutions Architect and Serverless evangelist with 20+ years in IT, holding multiple AWS certifications. At AWS, he drives customers through their cloud transformation journeys by converting complex challenges into actionable roadmaps for both technical and business audiences.

Rohit Chaudhari is a Senior Customer Solutions Manager with over 15 years of diverse tech experience. His background spans customer success, product management, digital transformation coaching, engineering, and consulting. At AWS, Rohit serves as a trusted advisor for customers to work backwards from their business goals, accelerate their journey to the cloud, and implement innovative solutions.

Al Destefano is a Generative AI Specialist at AWS based in New York City. Leveraging his AI/ML domain expertise, Al develops and executes global go-to-market strategies that drive transformative results for AWS customers at scale. He specializes in helping enterprise customers harness the power of Amazon Q, a generative AI-powered assistant, to overcome complex challenges and unlock new business opportunities.

Read More

The rise of large language models (LLMs) and foundation models (FMs) has revolutionized the field of natural language processing (NLP) and artificial intelligence (AI). These powerful models, trained on vast amounts of data, can generate human-like text, answer questions, and even engage in creative writing tasks. However, training and deploying such models from scratch is a complex and resource-intensive process, often requiring specialized expertise and significant computational resources.

Enter Amazon Bedrock, a fully managed service that provides developers with seamless access to cutting-edge FMs through simple APIs. Amazon Bedrock streamlines the integration of state-of-the-art generative AI capabilities for developers, offering pre-trained models that can be customized and deployed without the need for extensive model training from scratch. Amazon maintains the flexibility for model customization while simplifying the process, making it straightforward for developers to use cutting-edge generative AI technologies in their applications. With Amazon Bedrock, you can integrate advanced NLP features, such as language understanding, text generation, and question answering, into your applications.

In this post, we explore how to integrate Amazon Bedrock FMs into your code base, enabling you to build powerful AI-driven applications with ease. We guide you through the process of setting up the environment, creating the Amazon Bedrock client, prompting and wrapping code, invoking the models, and using various models and streaming invocations. By the end of this post, you’ll have the knowledge and tools to harness the power of Amazon Bedrock FMs, accelerating your product development timelines and empowering your applications with advanced AI capabilities.

Solution overview

Amazon Bedrock provides a simple and efficient way to use powerful FMs through APIs, without the need for training custom models. For this post, we run the code in a Jupyter notebook within VS Code and use Python. The process of integrating Amazon Bedrock into your code base involves the following steps:

1. Set up your development environment by importing the necessary dependencies and creating an Amazon Bedrock client. This client will serve as the entry point for interacting with Amazon Bedrock FMs.
2. After the Amazon Bedrock client is set up, you can define prompts or code snippets that will be used to interact with the FMs. These prompts can include natural language instructions or code snippets that the model will process and generate output based on.
3. With the prompts defined, you can invoke the Amazon Bedrock FM by passing the prompts to the client. Amazon Bedrock supports various models, each with its own strengths and capabilities, allowing you to choose the most suitable model for your use case.
4. Depending on the model and the prompts provided, Amazon Bedrock will generate output, which can include natural language text, code snippets, or a combination of both. You can then process and integrate this output into your application as needed.
5. For certain models and use cases, Amazon Bedrock supports streaming invocations, which allow you to interact with the model in real time. This can be particularly useful for conversational AI or interactive applications where you need to exchange multiple prompts and responses with the model.

Throughout this post, we provide detailed code examples and explanations for each step, helping you seamlessly integrate Amazon Bedrock FMs into your code base. By using these powerful models, you can enhance your applications with advanced NLP capabilities, accelerate your development process, and deliver innovative solutions to your users.

Prerequisites

Before you dive into the integration process, make sure you have the following prerequisites in place:

• AWS account – You’ll need an AWS account to access and use Amazon Bedrock. If you don’t have one, you can create a new account.
• Development environment – Set up an integrated development environment (IDE) with your preferred coding language and tools. You can interact with Amazon Bedrock using AWS SDKs available in Python, Java, Node.js, and more.
• AWS credentials – Configure your AWS credentials in your development environment to authenticate with AWS services. You can find instructions on how to do this in the AWS documentation for your chosen SDK. We walk through a Python example in this post.

With these prerequisites in place, you’re ready to start integrating Amazon Bedrock FMs into your code.

In your IDE, create a new file. For this example, we use a Jupyter notebook (Kernel: Python 3.12.0).

In the following sections, we demonstrate how to implement the solution in a Jupyter notebook.

Set up the environment

To begin, import the necessary dependencies for interacting with Amazon Bedrock. The following is an example of how you can do this in Python.

First step is to import boto3 and json:

import boto3, json

Next, create an instance of the Amazon Bedrock client. This client will serve as the entry point for interacting with the FMs. The following is a code example of how to create the client:

bedrock_runtime = boto3.client(
    service_name='bedrock-runtime',
    region_name='us-east-1'
)

Define prompts and code snippets

With the Amazon Bedrock client set up, define prompts and code snippets that will be used to interact with the FMs. These prompts can include natural language instructions or code snippets that the model will process and generate output based on.

In this example, we asked the model, “Hello, who are you?”.

To send the prompt to the API endpoint, you need some keyword arguments to pass in. You can get these arguments from the Amazon Bedrock console.

1. On the Amazon Bedrock console, choose Base models in the navigation pane.
   

2. Select Titan Text G1 – Express.
   

3. Choose the model name (Titan Text G1 – Express) and go to the API request.
   

4. Copy the API request:

{
"modelId": "amazon.titan-text-express-v1",
"contentType": "application/json",
"accept": "application/json",
"body": "{"inputText":"this is where you place your input text","textGenerationConfig":{"maxTokenCount":8192,"stopSequences":[],"temperature":0,"topP":1}}"
}

5. Insert this code in the Jupyter notebook with the following minor modifications:
   • We post the API requests to keyword arguments (kwargs).
   • The next change is on the prompt. We will replace ”this is where you place your input text” by ”Hello, who are you?”
6. Print the keyword arguments:

kwargs = {
 "modelId": "amazon.titan-text-express-v1",
 "contentType": "application/json",
 "accept": "application/json",
 "body": "{"inputText":"Hello, who are you?","textGenerationConfig":{"maxTokenCount":8192,"stopSequences":[],"temperature":0,"topP":1}}"
}
print(kwargs)

This should give you the following output:

{'modelId': 'amazon.titan-text-express-v1', 'contentType': 'application/json', 'accept': 'application/json', 'body': '{"inputText":"Hello, who are you?","textGenerationConfig":{"maxTokenCount":8192,"stopSequences":[],"temperature":0,"topP":1}}'}

Invoke the model

With the prompt defined, you can now invoke the Amazon Bedrock FM.

1. Pass the prompt to the client:

response = bedrock_runtime.invoke_model(**kwargs)
response

This will invoke the Amazon Bedrock model with the provided prompt and print the generated streaming body object response.

{'ResponseMetadata': {'RequestId': '3cfe2718-b018-4a50-94e3-59e2080c75a3',
'HTTPStatusCode': 200,
'HTTPHeaders': {'date': 'Fri, 18 Oct 2024 11:30:14 GMT',
'content-type': 'application/json',
'content-length': '255',
'connection': 'keep-alive',
'x-amzn-requestid': '3cfe2718-b018-4a50-94e3-59e2080c75a3',
'x-amzn-bedrock-invocation-latency': '1980',
'x-amzn-bedrock-output-token-count': '37',
'x-amzn-bedrock-input-token-count': '6'},
'RetryAttempts': 0},
'contentType': 'application/json',
'body': <botocore.response.StreamingBody at 0x105e8e7a0>}

The preceding Amazon Bedrock runtime invoke model will work for the FM you choose to invoke.

2. Unpack the JSON string as follows:

response_body = json.loads(response.get('body').read())
response_body

You should get a response as follows (this is the response we got from the Titan Text G1 – Express model for the prompt we supplied).

{'inputTextTokenCount': 6, 'results': [{'tokenCount': 37, 'outputText': 'nI am Amazon Titan, a large language model built by AWS. It is designed to assist you with tasks and answer any questions you may have. How may I help you?', 'completionReason': 'FINISH'}]}

Experiment with different models

Amazon Bedrock offers various FMs, each with its own strengths and capabilities. You can specify which model you want to use by passing the model_name parameter when creating the Amazon Bedrock client.

1. Like the previous Titan Text G1 – Express example, get the API request from the Amazon Bedrock console. This time, we use Anthropic’s Claude on Amazon Bedrock.
   

{
"modelId": "anthropic.claude-v2",
"contentType": "application/json",
"accept": "*/*",
"body": "{"prompt":"\n\nHuman: Hello world\n\nAssistant:","max_tokens_to_sample":300,"temperature":0.5,"top_k":250,"top_p":1,"stop_sequences":["\n\nHuman:"],"anthropic_version":"bedrock-2023-05-31"}"
}

Anthropic’s Claude accepts the prompt in a different way (\n\nHuman:), so the API request on the Amazon Bedrock console provides the prompt in the way that Anthropic’s Claude can accept.

2. Edit the API request and put it in the keyword argument:

   kwargs = {
     "modelId": "anthropic.claude-v2",
     "contentType": "application/json",
     "accept": "*/*",
     "body": "{"prompt":"\n\nHuman: we have received some text without any context.\nWe will need to label the text with a title so that others can quickly see what the text is about \n\nHere is the text between these <text></text> XML tags\n\n<text>\nToday I sent to the beach and saw a whale. I ate an ice-cream and swam in the sea\n</text>\n\nProvide title between <title></title> XML tags\n\nAssistant:","max_tokens_to_sample":300,"temperature":0.5,"top_k":250,"top_p":1,"stop_sequences":["\n\nHuman:"],"anthropic_version":"bedrock-2023-05-31"}"
   }
   print(kwargs)

You should get the following response:

{'modelId': 'anthropic.claude-v2', 'contentType': 'application/json', 'accept': '*/*', 'body': '{"prompt":"\n\nHuman: we have received some text without any context.\nWe will need to label the text with a title so that others can quickly see what the text is about \n\nHere is the text between these <text></text> XML tags\n\n<text>\nToday I sent to the beach and saw a whale. I ate an ice-cream and swam in the sea\n</text>\n\nProvide title between <title></title> XML tags\n\nAssistant:","max_tokens_to_sample":300,"temperature":0.5,"top_k":250,"top_p":1,"stop_sequences":["\n\nHuman:"],"anthropic_version":"bedrock-2023-05-31"}'}

3. With the prompt defined, you can now invoke the Amazon Bedrock FM by passing the prompt to the client:

response = bedrock_runtime.invoke_model(**kwargs)
response

You should get the following output:

{'ResponseMetadata': {'RequestId': '72d2b1c7-cbc8-42ed-9098-2b4eb41cd14e', 'HTTPStatusCode': 200, 'HTTPHeaders': {'date': 'Thu, 17 Oct 2024 15:07:23 GMT', 'content-type': 'application/json', 'content-length': '121', 'connection': 'keep-alive', 'x-amzn-requestid': '72d2b1c7-cbc8-42ed-9098-2b4eb41cd14e', 'x-amzn-bedrock-invocation-latency': '538', 'x-amzn-bedrock-output-token-count': '15', 'x-amzn-bedrock-input-token-count': '100'}, 'RetryAttempts': 0}, 'contentType': 'application/json', 'body': <botocore.response.StreamingBody at 0x1200b5990>}

4. Unpack the JSON string as follows:

response_body = json.loads(response.get('body').read())
response_body

This results in the following output on the title for the given text.

{'type': 'completion',
'completion': ' <title>A Day at the Beach</title>',
'stop_reason': 'stop_sequence',
'stop': 'nnHuman:'}

5. Print the completion:

completion = response_body.get('completion')
completion

Because the response is returned in the XML tags as you defined, you can consume the response and display it to the client.

' <title>A Day at the Beach</title>'

Invoke model with streaming code

For certain models and use cases, Amazon Bedrock supports streaming invocations, which allow you to interact with the model in real time. This can be particularly useful for conversational AI or interactive applications where you need to exchange multiple prompts and responses with the model. For example, if you’re asking the FM for an article or story, you might want to stream the output of the generated content.

1. Import the dependencies and create the Amazon Bedrock client:

import boto3, json
bedrock_runtime = boto3.client(
service_name='bedrock-runtime',
region_name='us-east-1'
)

2. Define the prompt as follows:

prompt = "write an article about fictional planet Foobar"

3. Edit the API request and put it in keyword argument as before:
   We use the API request of the claude-v2 model.

kwargs = {
  "modelId": "anthropic.claude-v2",
  "contentType": "application/json",
  "accept": "*/*",
  "body": "{"prompt":"\n\nHuman: " + prompt + "\nAssistant:","max_tokens_to_sample":300,"temperature":0.5,"top_k":250,"top_p":1,"stop_sequences":["\n\nHuman:"],"anthropic_version":"bedrock-2023-05-31"}"
}

4. You can now invoke the Amazon Bedrock FM by passing the prompt to the client:
   We use invoke_model_with_response_stream instead of invoke_model.

response = bedrock_runtime.invoke_model_with_response_stream(**kwargs)

stream = response.get('body')
if stream:
    for event in stream:
        chunk = event.get('chunk')
        if chunk:
            print(json.loads(chunk.get('bytes')).get('completion'), end="")

You get a response like the following as streaming output:

Here is a draft article about the fictional planet Foobar: Exploring the Mysteries of Planet Foobar Far off in a distant solar system lies the mysterious planet Foobar. This strange world has confounded scientists and explorers for centuries with its bizarre environments and alien lifeforms. Foobar is slightly larger than Earth and orbits a small, dim red star. From space, the planet appears rusty orange due to its sandy deserts and red rock formations. While the planet looks barren and dry at first glance, it actually contains a diverse array of ecosystems. The poles of Foobar are covered in icy tundra, home to resilient lichen-like plants and furry, six-legged mammals. Moving towards the equator, the tundra slowly gives way to rocky badlands dotted with scrubby vegetation. This arid zone contains ancient dried up riverbeds that point to a once lush environment. The heart of Foobar is dominated by expansive deserts of fine, deep red sand. These deserts experience scorching heat during the day but drop to freezing temperatures at night. Hardy cactus-like plants manage to thrive in this harsh landscape alongside tough reptilian creatures. Oases rich with palm-like trees can occasionally be found tucked away in hidden canyons. Scattered throughout Foobar are pockets of tropical jungles thriving along rivers and wetlands.

Conclusion

In this post, we showed how to integrate Amazon Bedrock FMs into your code base. With Amazon Bedrock, you can use state-of-the-art generative AI capabilities without the need for training custom models, accelerating your development process and enabling you to build powerful applications with advanced NLP features.

Whether you’re building a conversational AI assistant, a code generation tool, or another application that requires NLP capabilities, Amazon Bedrock provides a simple and efficient solution. By using the power of FMs through Amazon Bedrock APIs, you can focus on building innovative solutions and delivering value to your users, without worrying about the underlying complexities of language models.

As you continue to explore and integrate Amazon Bedrock into your projects, remember to stay up to date with the latest updates and features offered by the service. Additionally, consider exploring other AWS services and tools that can complement and enhance your AI-driven applications, such as Amazon SageMaker for machine learning model training and deployment, or Amazon Lex for building conversational interfaces.

To further explore the capabilities of Amazon Bedrock, refer to the following resources:

• Build custom generative AI applications powered by Amazon Bedrock
• Use Amazon Bedrock to generate, evaluate, and understand code in your software development pipeline
• Few-shot prompt engineering and fine-tuning for LLMs in Amazon Bedrock

Share and learn with our generative AI community at community.aws.

Happy coding and building with Amazon Bedrock!

About the Authors

Rajakumar Sampathkumar is a Principal Technical Account Manager at AWS, providing customer guidance on business-technology alignment and supporting the reinvention of their cloud operation models and processes. He is passionate about cloud and machine learning. Raj is also a machine learning specialist and works with AWS customers to design, deploy, and manage their AWS workloads and architectures.

YaduKishore Tatavarthi is a Senior Partner Solutions Architect at Amazon Web Services, supporting customers and partners worldwide. For the past 20 years, he has been helping customers build enterprise data strategies, advising them on Generative AI, cloud implementations, migrations, reference architecture creation, data modeling best practices, and data lake/warehouse architectures.

Read More

The emergence of generative AI has ushered in a new era of possibilities, enabling the creation of human-like text, images, code, and more. However, as exciting as these advancements are, data scientists often face challenges when it comes to developing UIs and to prototyping and interacting with their business users. Traditionally, building frontend and backend applications has required knowledge of web development frameworks and infrastructure management, which can be daunting for those with expertise primarily in data science and machine learning.

AWS provides a powerful set of tools and services that simplify the process of building and deploying generative AI applications, even for those with limited experience in frontend and backend development. In this post, we explore a practical solution that uses Streamlit, a Python library for building interactive data applications, and AWS services like Amazon Elastic Container Service (Amazon ECS), Amazon Cognito, and the AWS Cloud Development Kit (AWS CDK) to create a user-friendly generative AI application with authentication and deployment.

Solution overview

For this solution, you deploy a demo application that provides a clean and intuitive UI for interacting with a generative AI model, as illustrated in the following screenshot.

The UI consists of a text input area where users can enter their queries, and an output area to display the generated results.

The default interface is simple and straightforward, but you can extend and customize it to fit your specific needs. With Streamlit’s flexibility, you can add additional features, adjust the styling, and integrate other functionalities as required by your use case.

The solution we explore consists of two main components: a Python application for the UI and an AWS deployment architecture for hosting and serving the application securely.

The Python application uses the Streamlit library to provide a user-friendly interface for interacting with a generative AI model. Streamlit allows data scientists to create interactive web applications using Python, using their existing skills and knowledge. With Streamlit, you can quickly build and iterate on your application without the need for extensive frontend development experience.

The AWS deployment architecture makes sure the Python application is hosted and accessible from the internet to authenticated users. The solution uses the following key components:

• Amazon ECS and AWS Fargate provide a serverless container orchestration platform for running the Python application
• Amazon Cognito handles user authentication, making sure only authorized users can access the generative AI application
• Application Load Balancer (ALB) and Amazon CloudFront are responsible for load balancing and content delivery, so the application is available for users worldwide
• The AWS CDK allows you to define and provision AWS infrastructure resources using familiar programming languages like Python
• Amazon Bedrock is a fully managed service that offers a choice of high-performing generative AI models through an API

The following diagram illustrates this architecture.

Prerequisites

As a prerequisite, you need to enable model access in Amazon Bedrock and have access to a Linux or macOS development environment. You could also use a Windows development environment, in which case you need to update the instructions in this post.

Access to Amazon Bedrock foundation models is not granted by default. Complete the following steps to enable access to Anthropic’s Claude on Amazon Bedrock, which we use as part of this post:

1. Sign in to the AWS Management Console.
2. Choose the us-east-1 AWS Region from the top right corner.
3. On the Amazon Bedrock console, choose Model access in the navigation pane.
4. Choose Manage model access.
5. Select the model you want access to (for this post, Anthropic’s Claude). You can also select other models for future use.
6. Choose Next and then Submit to confirm your selection.

For more information on how to manage model access, see Access Amazon Bedrock foundation models.

Set up your development environment

To get started with deploying the Streamlit application, you need access to a development environment with the following software installed:

• Python version 3.8 or later. You can download Python from the official website or use your Linux distribution’s package manager.
• The AWS Command Line Interface (AWS CLI). For installation instructions, see Installing or updating to the latest version of the AWS CLI.
• The AWS CDK. For installation instructions, see Getting started with the AWS CDK.
• Docker or Colima. For installation instructions, see Install Docker Engine. For macOS, we have tested the deployment with Colima container runtimes in replacement for Docker Desktop.

You also need to configure the AWS CLI. One way to do it is to get your access key through the console, and use the aws configure command in your terminal to set up your credentials.

Clone the GitHub repository

Use the terminal of your development environment to enter the commands in the following steps:

1. Clone the deploy-streamlit-app repository from the AWS Samples GitHub repository:

git clone https://github.com/aws-samples/deploy-streamlit-app.git

2. Navigate to the cloned repository:

cd deploy-streamlit-app

Create the Python virtual environment and install the AWS CDK

Complete the following steps to set up the virtual environment and the AWS CDK:

1. Create a new Python virtual environment (your Python version should be 3.8 or greater):

python3 -m venv .venv

2. Activate the virtual environment:

source .venv/bin/activate

3. Install the AWS CDK, which is in the required Python dependencies:

pip install -r requirements.txt

Configure the Streamlit application

Complete the following steps to configure the Streamlit application:

1. In the docker_app directory, locate the config_file.py file.
2. Open config_file.py in your editor and modify the STACK_NAME and CUSTOM_HEADER_VALUE variables:
   1. The stack name enables you to deploy multiple applications in the same account. Choose a different stack name for each application. For your first application, you can leave the default value.
   2. The custom header value is a security token that CloudFront uses to authenticate on the load balancer. You can choose it randomly, and it must be kept secret.

Deploy the AWS CDK template

Complete the following steps to deploy the AWS CDK template:

1. From your terminal, bootstrap the AWS CDK:

cdk bootstrap

2. Deploy the AWS CDK template, which will create the necessary AWS resources:

cdk deploy

3. Enter y (yes) when asked if you want to deploy the changes.

The deployment process may take 5–10 minutes. When it’s complete, note the CloudFront distribution URL and Amazon Cognito user pool ID from the output.

Create an Amazon Cognito user

Complete the following steps to create an Amazon Cognito user:

1. On the Amazon Cognito console, navigate to the user pool that you created as part of the AWS CDK deployment.
2. On the Users tab, choose Create user.

3. Enter a user name and password.
4. Choose Create user.

Access the Streamlit application

Complete the following steps to access the Streamlit application:

1. Open a new web browser window or tab and navigate to the CloudFront distribution URL from the AWS CDK deployment output.

If you have not noted this URL, you can open the AWS CloudFormation console and find it in the outputs of the stack.

2. Log in to the Streamlit application using the Amazon Cognito user credentials you created in the previous step.

You should now be able to access and interact with the Streamlit application, which is deployed and running on AWS using the provided AWS CDK template.

This deployment is intended as a starting point and a demo. Before using this application in a production environment, you should thoroughly review and implement appropriate security measures, such as configuring HTTPS on the load balancer and following AWS best practices for securing your resources. See the README.md file in the GitHub repository for more information.

Customize the application

The aws-samples/deploy-streamlit-app GitHub repository provides a solid foundation for building and deploying generative AI applications, but it’s also highly customizable and extensible.

Let’s explore how you can customize the Streamlit application. Because the application is written in Python, you can modify it to integrate with different generative AI models, add new features, or change the UI to better align with your application’s requirements.

For example, let’s say you want to add a button to invoke the LLM answer instead of invoking it automatically when the user enters input text. Complete the following steps to modify the docker_app/app.py file:

1. After the definition of the input_sent text input, add a Streamlit button:

# Insert this after the line starting with input_sent = …
submit_button = st.button("Get LLM Response")

2. Change the if condition to check if the button is clicked instead of checking for input_sent:

# Replace the line `if input_sent:` by the following
if submit_button:

3. Redeploy the application by entering the following in the terminal:

cdk deploy

The deployment should take less than 5 minutes. In the next section, we show how to test your changes locally before deploying, which will accelerate your development workflow.

4. When the deployment is complete, refresh the webpage in your browser.

The Streamlit application will now display a button labeled Get LLM Response. When the user chooses this button, the LLM will be invoked, and the output will be displayed on the UI.

This is just one example of how you can customize the Streamlit application to meet your specific requirements. You can modify the code further to integrate with different generative AI models, add additional features, or enhance the UI as needed.

Test your changes locally before deploying

Although deploying the application using cdk deploy allows you to test your changes in the actual AWS environment, it can be time-consuming, especially during the development and testing phase. Fortunately, you can run and test your application locally before deploying it to AWS.

To test your changes locally, follow these steps:

1. In your terminal, navigate to the docker_app directory, where the Streamlit application is located:

cd docker_app

2. If you haven’t already, install the dependencies of the Python application. These dependencies are different from the ones of the AWS CDK application that you installed previously.

pip install -r requirements.txt

3. Start the Streamlit server with the following command:

streamlit run app.py --server.port 8080

This will start the Streamlit application on port 8080.

You should now be able to interact with the locally running Streamlit application and test your changes without having to redeploy the application to AWS.

Remember to stop the Streamlit server (by pressing Ctrl+C in the terminal) when you’re done testing.

By testing your changes locally, you can significantly speed up the development and testing cycle, allowing you to iterate more quickly and catch issues early in the process.

Clean up

To avoid incurring additional charges, clean up the resources created during this demo:

1. Open the terminal in your development environment.
2. Make sure you’re in the root directory of the project and your virtual environment is activated:

cd ~/environment/deploy-streamlit-app
source .venv/bin/activate

3. Destroy the AWS CDK stack:

cdk destroy

4. Confirm the deletion by entering yes when prompted.

Conclusion

Building and deploying user-friendly generative AI applications no longer requires extensive knowledge of frontend and backend development frameworks. By using Streamlit and AWS services, data scientists can focus on their core expertise while still delivering secure, scalable, and accessible applications to business users.

The full code of the demo is available in the GitHub repository. It provides a valuable starting point for building and deploying generative AI applications, allowing you to quickly set up a working prototype and iterate from there. We encourage you to explore the repository and experiment with the provided solution to create your own applications.

As the adoption of generative AI continues to grow, the ability to build and deploy user-friendly applications will become increasingly important. With AWS and Python, data scientists now have the tools and resources to bridge the gap between their technical expertise and the need to showcase their models to business users through secure and accessible UIs.

About the Author

Lior Perez is a Principal Solutions Architect on the Construction team based in Toulouse, France. He enjoys supporting customers in their digital transformation journey, using big data, machine learning, and generative AI to help solve their business challenges. He is also personally passionate about robotics and IoT, and constantly looks for new ways to use technologies for innovation.

Read More

Generative AI continues to push the boundaries of what’s possible. One area garnering significant attention is the use of generative AI to analyze audio and video transcripts, increasing our ability to extract valuable insights from content stored in audio or video files. Speech data is unique and complex, which makes it difficult to analyze and extract insights. Manually transcribing and analyzing it can be time-consuming and resource-intensive.

Existing methods for extracting insights from speech data often require tedious human transcription and review. You can use automatic voice recognition tools to convert your audio and video data to text. However, you still have to rely on manual processes for extracting specific insights and data points, or get summaries of the content. This approach is time-consuming and as organizations amass vast amounts of this content, the need for a more efficient and insightful solution becomes increasingly pressing. There is a significant opportunity to add business value given the amount of data organizations store in these formats and the valuable insights that might otherwise go undiscovered. The following are some of the new insights and capabilities that can be obtained through the use of large language models (LLM) with audio transcripts:

• LLMs can analyze and understand the context of a conversation, not just the words spoken, but also the implied meaning, intent, and emotions. Previously, this would have required extensive human interpretation.
• LLMs can perform advanced sentiment analysis. Previously, sentiment analysis could be captured, but LLMs can capture more emotions, such as sarcasm, ambivalence, or mixed feelings by understanding the context of the conversation.
• LLMs can generate concise summarizations not just by extracting content, but by understanding the context of the conversation.
• Users can now ask complex, natural language questions and receive insightful answers.
• LLMs can infer personas or roles in a conversation, enabling targeted insights and actions.
• LLMs can support the creation of new content based on audio assets or conversations following predetermined templates or flows.

In this post, we examine how to create business value through speech analytics with some examples focused on the following:

• Automatically summarizing, categorizing, and analyzing marketing content such as podcasts, recorded interviews, or videos, and creating new marketing materials based on those assets
• Automatically extracting key points, summaries, and sentiment from a recorded meeting (such as an earnings call)
• Transcribing and analyzing contact center calls to improve customer experience.

The first step in getting these audio data insights involves transcribing the audio file using Amazon Transcribe. Amazon Transcribe is a machine learning (ML) based managed service that automatically converts speech to text, enabling developers to seamlessly integrate speech-to-text capabilities into their applications. It also recognizes multiple speakers, automatically redacts personally identifiable information (PII), and allows you to enhance the accuracy of a transcription by providing custom vocabularies specific to your industries or use case, or by using custom language models.

The second step involves using foundation models (FMs) with Amazon Bedrock to summarize the content, identify topics, and recognize conclusions, extracting valuable insights that can guide strategic decisions and innovations. Automatic generation of new content also adds value, increasing creativity and productivity.

Generative AI is reshaping the way we analyze audio transcripts, enabling you to unlock insights such as customer sentiment, pain points, common themes, avenues for risk mitigation, and more, that were previously obfuscated.

Use case overview

In this post, we discuss three example use cases in detail. The code artifacts are in Python. We used a Jupyter notebook to run the code snippets. You can follow along by creating and running a notebook in Amazon SageMaker Studio.

Audio summarization and insights, and automated generation of new content using Amazon Transcribe and Amazon Bedrock

Through this use case, we demonstrate how to take an existing marketing asset (a video) and create a new blog post to announce the launch of the video, create an abstract, and extract the main topics and the search engine optimization (SEO) keywords present in the post for documenting and categorizing the asset.

Transcribe audio with Amazon Transcribe

In this case, we use an AWS re:Invent 2023 technical talk as a sample. For the purpose of this notebook, we downloaded the MP4 file for the recording and stored it in an Amazon Simple Storage Service (Amazon S3) bucket.

The first step is to transcribe the audio file using Amazon Transcribe:

# Create a Amazon Transcribe transcirption job by specifying the audio/video file's S3 location
import boto3
import time
import random
transcribe = boto3.client('transcribe')
response = transcribe.start_transcription_job(
    TranscriptionJobName=f"podcast-transcription-{int(time.time())}_{random.randint(1000, 9999)}",
    LanguageCode='en-US',
    MediaFormat='mp3',
    Media={
        'MediaFileUri': '<S3 URI of the media file>'
    },
    OutputBucketName='<name of the S3 bucket that will store the output>',
    OutputKey='transcribe_bedrock_blog/output-files/',
    Settings={
        'ShowSpeakerLabels': True,
        'MaxSpeakerLabels': 3
    }
)
max_tries = 60
while max_tries > 0:
    max_tries -= 1
    job = transcribe.get_transcription_job(TranscriptionJobName=response['TranscriptionJob']['TranscriptionJobName'])
    job_status = job["TranscriptionJob"]["TranscriptionJobStatus"]
    if job_status in ["COMPLETED", "FAILED"]:
        if job_status == "COMPLETED":
            print(
                f"Download the transcript fromn"
                f"t{job['TranscriptionJob']['Transcript']['TranscriptFileUri']}."
            )
        break
    else:
        print(f"Waiting for {response['TranscriptionJob']['TranscriptionJobName']}. Current status is {job_status}.")
    time.sleep(10)

The transcription job will take a few minutes to complete.

When the job is complete, you can inspect the transcription output and check the plain text transcript that was generated (the following has been trimmed for brevity):

# Get the Transcribe Output JSON file
s3 = boto3.client('s3')
output_bucket = job['TranscriptionJob']['Transcript']['TranscriptFileUri'].split('https://')[1].split('/',2)[1]
output_file_key = job['TranscriptionJob']['Transcript']['TranscriptFileUri'].split('https://')[1].split('/',2)[2]
s3_response_object = s3.get_object(Bucket=output_bucket, Key=output_file_key)
object_content = s3_response_object['Body'].read()

transcription_output = json.loads(object_content)

# Let's see what we have inside the job output JSON
print(transcription_output['results']['transcripts'][0]['transcript'])

……….Once the alert comes, how do you kind of correlate these alerts, not by just text signals and text passing, but understanding the topology, the infrastructure topology that is supporting that application or that business service. It is the topology that ultimately gives the source of truth when an alert comes, right. That’s what we mean by a correlation that is assisted with topology in in in the thing that ultimately results in finding a probable root cause. And once ……….

After you have validated the existence of the text, you can use Amazon Bedrock to analyze the output:

# Now let's use this transcript to extract insights with a help of a Large Language Model on Amazon Bedrock
# First let's initialize the bedrock runtime client to invoke the model. 
bedrock_runtime = boto3.client('bedrock-runtime')
# Selecting Claude 3 Sonnet
model_id = 'anthropic.claude-3-sonnet-20240229-v1:0'

Using the transcription from the technical talk, we use Amazon Bedrock to call an FM (we use Anthropic’s Claude 3 Sonnet on Amazon Bedrock in this case). You can choose from the language models available on Amazon Bedrock from AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon.

You can now perform additional tasks.

Extract the main topics with Amazon Bedrock

The following prompt provides instructions to ask the LLM for the main topics in the technical talk:

# Extracting the main topics with Amazon Bedrock
main_topics_prompt = """Based on the contents of <transcript></transcript>, what are the main topics being discussed? Display the topics as a list.

<transcript>
{transcript}
</transcript>
"""

user_message = {"role": "user", "content": main_topics_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
{
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1000,
"messages": messages
}
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
main_topics = response_body['content'][0]['text']
print(main_topics)

We have created a prompt that uses prompting best practices for Anthropic’s Claude. In this case, we pass the transcript within the <transcript></transcript> XML tags and ask for main topics discussed. Based on that, we get the following output:

Based on the contents of the transcript, the main topics being discussed are:

1. AI and Machine Learning in IT Operations
2. Challenges faced by IT Operations teams
3. Observable data explosion from cloud, on-premise, and edge infrastructure
4. The convergence of IT, SRE, and DevOps roles
5. Discovery to Resolution pipeline for IT Operations
6. Alert correlation and intelligent alerting using AI/ML
7. Probable root cause analysis with AI/ML
8. Automated remediation and runbook automation
9. Predictive analytics and preventive maintenance using AI/ML
10. Benefits of AI/ML for IT Operations (proactive operations, reduced human effort, improved service health, cost optimization)

Generate a blog post to announce the video with Amazon Bedrock

The following prompt provides instructions to ask the LLM for the generation of a new blog post announcing the launch of the preceding recording. Just like the preceding prompt, we use prompting best practices for Anthropic’s Claude:

# Generating a blog post to announce the Podcast with Amazon Bedrock
generate_blog_prompt = """The contents of <transcript></transcript> are a transcription of a technical talk video recording. Based on this information write a blogpost that summarizes the technical talk and announces that the recording of this talk is now available for public consumption. Limit the blogpost to 1500 words. Do not include any information that is not mentioned in <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": generate_blog_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
generate_blog = response_body['content'][0]['text']
print(generate_blog)

Based on that, we get the following output:

Here is a draft 1462 word blogpost summarizing the technical talk and announcing the recording is now available:

Title: Revolutionizing IT Operations with AI/ML: A Future-Ready Approach

The world of IT operations is undergoing a seismic shift, driven by the explosion of observable data, the convergence of IT roles, and the advent of hybrid multi-cloud environments. Traditional reactive approaches are no longer sufficient to keep pace with the complexity and scale of modern IT systems. This is where AI and machine learning (ML) come into play, offering a future-ready approach to revolutionize IT operations.

In a recent technical talk, Verma from HPE OpsRamp outlined the drivers behind this AI/ML imperative and demonstrated how this cutting-edge technology is fundamentally transforming IT operations. The recording of this insightful session is now available for public consumption.

The Drivers Behind the AI/ML Imperative

The first driving force behind the adoption of AI/ML in IT operations is the sheer volume of observable data that………………..

Create an abstract with Amazon Bedrock

The following prompt provides instructions to create an abstract summarizing the content of the technical talk:

# Creating an Abstract with Amazon Bedrock
abstract_prompt = """The contents of <transcript></transcript> are a transcription of a technical talk video recording. Based on this information write an abstract for this technical talk in no more than 100 words. Do not include any information that is not mentioned in <transcript></transcript>.

<transcript>
{transcript}
</transcript>
"""

user_message = {"role": "user", "content": abstract_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
{
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1000,
"messages": messages
}
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
abstract = response_body['content'][0]['text']
print(abstract)

Based on this, we get the following output:

Here is a 96 word abstract for the technical talk:

This talk discussed how AI and machine learning innovations are transforming IT operations for hybrid multi-cloud workloads. The speakers outlined the challenges of dealing with exploding observable data from modern infrastructure and the convergence of IT roles. They demonstrated how AI/ML techniques like intelligent alerting, alert correlation, probable root cause analysis, and automated remediation can drive more proactive, predictive operations. Key benefits showcased included reduced human effort, improved service health, and cost optimization. The talk featured a demo of the OpsRamp platform leveraging AI/ML models to streamline the discovery-to-resolution pipeline for managing edge-to-cloud environments.

Extract SEO keywords from the generated blog post with Amazon Bedrock

The following prompt provides instructions to extract the main SEO keywords from the generated blog post. Based on online research on how to extract SEO keywords from long-form text, we came up with the following list. This demonstrates how you can empower an LLM like Anthropic’s Claude to follow instructions and best practices for a particular task or domain. Also, the prompt specifies that the output should be in JSON. This is helpful in use cases where you want to programmatically get results from an LLM and therefore require consistent formatting. Based on best practices for Anthropic’s Claude, we use the Assistant message in the messages API to pre-fill the model’s response to have further control on the output format:

# Extracting SEO keywords from the generated blog post
SEO_keywords_prompt = """Extract the most relevant keywords and phrases from the given blog post text present in <blog></blog> that would be valuable for SEO (search engine optimization) based on the instructions present in <instructions></instructions> below. The ideal keywords should capture the main topics, concepts, entities, and high-value terms present in the content. Use JSON format with key "keywords" and value as an array of keywords. Skip the preamble; go straight into the JSON. 

<blog>
{textblog}
</blog>

<instructions>
1. Carefully read through the entire blog post text to understand the main topics, concepts, and ideas covered.
2. Identify important nouns, noun phrases, multi-word phrases, and relevant adjective-noun combinations that relate to the core subject matter of the post.
3. Look for words and phrases that potential searchers might use to find content like this.
4. Prioritize terms that are highly specific and relevant to the blog topic over generic words.
5. Vary the keyword length and include both head terms (shorter, more popular keywords) and long-tail terms (longer, more specific phrases).
6. Aim to extract around 10-20 of the most valuable, high-impact keywords and phrases for SEO.
</instructions>
"""

user_message =  {"role": "user", "content": SEO_keywords_prompt.format(textblog = generate_blog)}
assistant_message = {"role": "assistant", "content": '{"keywords": ['}
messages = [user_message, assistant_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
SEO_keywords = response_body['content'][0]['text']
SEO_keywords = '{"keywords": [' + SEO_keywords
SEO_keywords_json = json.loads(SEO_keywords)
print(SEO_keywords_json)

Based on this, we get the following output:

{‘keywords’: [‘AI-driven IT operations’, ‘machine learning IT operations’, ‘proactive IT operations’, ‘predictive IT operations’, ‘AI for hybrid cloud’, ‘AI for multi-cloud’, ‘FutureOps’, ‘AI-assisted IT operations’, ‘AI-powered event correlation’, ‘intelligent alerting’, ‘automated remediation workflows’, ‘predictive analytics for IT’, ‘AI anomaly detection’, ‘AI root cause analysis’, ‘AI-driven observability’, ‘AI for DevOps’, ‘AI for SRE’, ‘AI IT operations management’]}

For consistent formatting and structured output, you can also use the Converse and ConverseStream APIs in Amazon Bedrock and use the tool calling capabilities of the LLMs that offer it.

Generate a new blog post version emphasizing specific SEO keywords with Amazon Bedrock

The following prompt provides instructions to generate a new blog post emphasizing three keywords or phrases from the ones extracted in the previous step. In this case, we use the first three keywords in the returned keywords list. The instructions specified in the prompt are derived from SEO content best practices:

# Extracting SEO keywords from the generated blog post
rewrite_SEO_keywords_prompt = """Rewrite the given blog post text present in <blog></blog> to improve its ranking for 3 target keywords listed in <keywords></keywords>. Rewrite the blog post, naturally incorporating the target keywords multiple times throughout the text. Keep the rewritten content high-quality, readable, and relevant, while strategically emphasizing the keywords to improve SEO.

<blog>
{textblog}
</blog>

<keywords>
{kw1},
{kw2},
{kw3}
</keywords>
"""

user_message = {"role": "user", "content": rewrite_SEO_keywords_prompt.format(textblog = generate_blog, kw1 = SEO_keywords_json['keywords'][0], kw2 = SEO_keywords_json['keywords'][1], kw3 = SEO_keywords_json['keywords'][2])}
messages = [user_message]

body=json.dumps(
{
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1000,
"messages": messages
}
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
rewrite_SEO_blog = response_body['content'][0]['text']
print(rewrite_SEO_blog)

Based on this, we get the following output:

<blog>

Title: Unleash the Power of AI-Driven and Machine Learning IT Operations for Proactive IT Operations

In today’s rapidly evolving digital landscape, traditional IT operations approaches are struggling to keep pace. The explosion of data from cloud workloads, Kubernetes clusters, and distributed edge locations has overwhelmed conventional monitoring tools, leading to fragmented views and reactive operations that respond to issues after they occur. To stay ahead of the curve, organizations must embrace AI-driven and machine learning IT operations, paving the way for proactive IT operations that enhance efficiency and minimize downtime.

At a recent technical talk, Verma and Radha Krishna Kunni from OpsRamp (recently acquired by HPE) delved into the transformative impact of AI and machine learning on IT operations, DevOps, and SRE for hybrid multi-cloud environments. They highlighted the key challenges ops teams face today and introduced the innovative “FutureOps” approach, which leverages AI and machine learning to revolutionize IT operations.

The full video recording of this insightful technical talk is now available [link], providing a comprehensive understanding of…………

Summarize content discussed in a recorded meeting using Amazon Transcribe and Amazon Bedrock

Through this use case, we demonstrate how to take an existing recording from a meeting (we use a recording from an AWS earnings call) to summarize the content discussed, extract the key points, and provide details on the sentiment of the meeting. For additional information on this use case, see Live Meeting Assistant with Amazon Transcribe, Amazon Bedrock, and Amazon Bedrock Knowledge Bases or Amazon Q Business.

Transcribe audio with Amazon Transcribe

In this use case, we use an Amazon 2024 Q1 earnings call as a sample. For the purpose of this notebook, we downloaded the WAV file for the recording and stored in an S3 bucket.

The first step is to transcribe the audio file using Amazon Transcribe:

# Create a Amazon Transcribe transcription job by specifying the audio/video file's S3 location
import boto3
import time
import random
transcribe = boto3.client('transcribe')
response = transcribe.start_transcription_job(
    TranscriptionJobName=f"meeting-transcription-{int(time.time())}_{random.randint(1000, 9999)}",
    LanguageCode='en-US',
    MediaFormat='mp3',
    Media={
        'MediaFileUri': '<S3 URI of the media file>'
    },
    OutputBucketName='<name of the S3 bucket that will store the output>',
    OutputKey='transcribe_bedrock_blog/output-files/',
    Settings={
        'ShowSpeakerLabels': True,
        'MaxSpeakerLabels': 10
    }
)
# Check whether the transcribe job is complete

max_tries = 60
while max_tries > 0:
    max_tries -= 1
    job = transcribe.get_transcription_job(TranscriptionJobName=response['TranscriptionJob']['TranscriptionJobName'])
    job_status = job["TranscriptionJob"]["TranscriptionJobStatus"]
    if job_status in ["COMPLETED", "FAILED"]:
        if job_status == "COMPLETED":
            print(
                f"Download the transcript fromn"
                f"t{job['TranscriptionJob']['Transcript']['TranscriptFileUri']}."
            )
        break
    else:
        print(f"Waiting for {response['TranscriptionJob']['TranscriptionJobName']}. Current status is {job_status}.")
    time.sleep(10)

The transcription job will take a few minutes to complete.

When the job is complete, you can inspect the transcription output and check for the plain text transcript that was generated:

import json
# Get the Transcribe Output JSON file
s3 = boto3.client('s3')
output_bucket = job['TranscriptionJob']['Transcript']['TranscriptFileUri'].split('https://')[1].split('/',2)[1]
output_file_key = job['TranscriptionJob']['Transcript']['TranscriptFileUri'].split('https://')[1].split('/',2)[2]
s3_response_object = s3.get_object(Bucket=output_bucket, Key=output_file_key)
object_content = s3_response_object['Body'].read()

transcription_output = json.loads(object_content)

# Let's see what we have inside the job output JSON
print(transcription_output['results']['transcripts'][0]['transcript'])

Thank you for standing by. Good day, everyone and welcome to the amazon.com first quarter, 2024 financial results teleconference. At this time, all participants are in a listen only mode. After the presentation, we will conduct a question and answer session. Today’s call is being recorded and for opening remarks, I’ll be turning the call over to the Vice President of Investor………

After you have validated the existence of the text, you can use Amazon Bedrock to analyze the output:

# Now let's use this transcript to extract insights with a help of a Large Language Model on Amazon Bedrock
# First let's initialize the bedrock runtime client to invoke the model. 
bedrock_runtime = boto3.client('bedrock-runtime')
# Selecting Claude 3 Sonnet
model_id = 'anthropic.claude-3-sonnet-20240229-v1:0'

Using the transcription from the earnings call recording, we use Amazon Bedrock to call an FM (we use Anthropic’s Claude 3 Sonnet in this case). You can choose from other FMs available on Amazon Bedrock.

You can now perform additional tasks.

Identify the financial ratios highlighted during this earnings call

The following prompt provides instructions to identify financial ratios highlighted during the earnings call and their implications:

# Identify the financial ratios highlighted during this earnings call 
financial_ratios_prompt = """Based on the contents of <transcript></transcript>,identify the financial ratios highlighted during this earnings call and their implications <transcript></transcript> . 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": financial_ratios_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
financial_ratios = response_body['content'][0]['text']
print(financial_ratios)

Based on this, we get the following output:

Based on the earnings call transcript, the following financial ratios and their implications were highlighted:

1. Operating Income Margin:
   • Amazon reported its highest ever quarterly operating income of $15.3 billion, which was $3.3 billion above the high end of their guidance range. This was driven by strong operational performance across all three reportable segments (North America, International, and AWS) and better-than-expected operating leverage, including lower cost to serve.
   • North America segment operating income was $5 billion with an operating margin of 5.8%, up 460 basis points year-over-year, driven by improvements in cost to serve, including benefits from regionalization efforts, more consolidated customer shipments, and improved leverage.
   • International segment operating income was $903 million with an operating margin of 2.8%, up 710 basis points year-over-year, primarily driven by cost efficiencies through network design enhancements and improved volume leverage in established countries, as well as progress in emerging countries.
   • AWS operating income was $9.4 billion, an increase of $4.3 billion year-over-year, with improved leverage from managing infrastructure and fixed costs while growing at a healthy rate.

Implication: The higher operating income margins across all segments indicate Amazon’s focus on driving efficiencies and improving profitability while continuing to invest in growth opportunities.

2. Revenue Growth:
   • Worldwide revenue was $143.3 billion, up 13% year-over-year (excluding the impact of foreign exchange).
   • AWS revenue grew 17.2% year-over-year, accelerating from 13.2% in Q4 2023, driven by strong demand for both generative AI and non-generative AI workloads.
   • Advertising revenue grew 24% year-over-year (excluding the impact of foreign exchange), primarily driven by sponsored products and improvements in relevancy and measurement capabilities.

Implication: The strong revenue growth, particularly in AWS and advertising, highlights Amazon’s diversified revenue streams and the growth opportunities in cloud computing and digital advertising.

3. Capital Expenditures (Capex):
   • Amazon anticipates a meaningful increase in overall capital expenditures in 2024, primarily driven by higher infrastructure Capex for growth in AWS, including generative AI investments.
   • In Q1 2024, Capex was $14 billion, expected to be the lowest quarter of the year.

Implication: The increase in Capex signals Amazon’s confidence in the strong demand for AWS and their commitment to investing in emerging technologies like generative AI to drive future growth.

Overall, the financial ratios and commentary indicate Amazon’s focus on improving profitability, driving operational efficiencies, and investing in growth opportunities, particularly in AWS and generative AI, while maintaining a diversified revenue stream and managing costs effectively.

Identify the speakers from the earnings call with Amazon Bedrock

The following prompt provides instructions to identify the speakers in the meeting from the transcription:

# Identify the speakers from the earnings call 
speakers_prompt = """Based on the contents of <transcript></transcript>,identify the speakers on this earnings call <transcript></transcript> . 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": speakers_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
speakers = response_body['content'][0]['text']
print(speakers)

Based on this, we get the following output:

Based on the transcript, the key speakers on this Amazon earnings call appear to be:

1. Andy Jassy – CEO of Amazon
2. Brian Olsavsky – CFO of Amazon
3. Dave Fildes – Vice President of Investor Relations at Amazon

The call begins with opening remarks from Dave Fildes, followed by prepared statements from Andy Jassy and Brian Olsavsky. They then take questions from analysts, with Andy and Brian providing the responses.

Obtain the challenges or negative areas discussed on the earnings call with Amazon Bedrock

The following prompt provides instructions to obtain the challenges or negative areas discussed from the transcription:

# Obtain the challenges or negative areas discussed on earnings call

challenges_prompt = """Based on the contents of <transcript></transcript>,Obtaining the challenges or negative areas discussed on earnings <transcript></transcript> . 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": challenges_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
challenges = response_body['content'][0]['text']
print(challenges)

Based on this, we get the following output:

Based on the transcript, some of the key challenges or negative areas discussed include:

1. Foreign exchange headwinds: Amazon faced an unfavorable impact from global currencies weakening against the U.S. dollar in Q1, leading to a $700 million or 50 basis point headwind to revenue compared to guidance.
2. Increasing capital expenditures: Amazon expects to meaningfully increase its capital expenditures year-over-year in 2024, primarily driven by higher infrastructure spending for AWS growth, including investments in generative AI capabilities.
3. Consumer spending concerns: Amazon mentioned keeping an eye on consumer spending trends, specifically in Europe, where it appears weaker relative to the U.S.
4. International segment profitability: While the international segment’s profitability improved, with an operating margin of 2.8%, Amazon acknowledged the need to continue working on cost efficiencies and profitability, particularly in emerging countries.
5. Cost optimization challenges: Although Amazon believes the majority of cost optimization efforts are behind them, there is still a need to continually streamline processes, optimize inventory placement, and invest in automation to further reduce the cost to serve.

Overall, the challenges centered around foreign exchange impacts, increasing capital intensity for AWS and generative AI investments, consumer demand uncertainties, and ongoing efforts to improve operational efficiencies and international profitability.

Get insights from a call center call between an agent and a customer using Amazon Transcribe and Amazon Bedrock

Through this use case, we demonstrate how to take an existing call recording from a contact center and summarize the content discussed, extract the main topic, key phrases, call reason, customer satisfaction, overall call sentiment, and sentiment about the products and services discussed. For additional details about this use case, see Live call analytics and agent assist for your contact center with Amazon language AI services and Post call analytics for your contact center with Amazon language AI services.

Transcribe audio with Amazon Transcribe

The first step is to transcribe the audio file using Amazon Transcribe. In this case, we use a sample from the Amazon Transcribe Post Call Analytics Solution GitHub repository. For the purpose of this notebook, we downloaded the WAV file and stored it in an S3 bucket.

# Create a Amazon Transcribe transcirption job by specifying the audio/video file's S3 location
import boto3
import time
import random
transcribe = boto3.client('transcribe')
response = transcribe.start_transcription_job(
    TranscriptionJobName=f"call_center-transcription-{int(time.time())}_{random.randint(1000, 9999)}",
    LanguageCode='en-US',
    MediaFormat='mp3',
    Media={
        'MediaFileUri': '<S3 URI of the media file>'
    },
    OutputBucketName='<name of the S3 bucket that will store the output>',
    OutputKey='transcribe_bedrock_blog/output-files/',
    Settings={
        'ShowSpeakerLabels': True,
        'MaxSpeakerLabels': 3
    }
)
max_tries = 60
while max_tries > 0:
    max_tries -= 1
    job = transcribe.get_transcription_job(TranscriptionJobName=response['TranscriptionJob']['TranscriptionJobName'])
    job_status = job["TranscriptionJob"]["TranscriptionJobStatus"]
    if job_status in ["COMPLETED", "FAILED"]:
        if job_status == "COMPLETED":
            print(
                f"Download the transcript fromn"
                f"t{job['TranscriptionJob']['Transcript']['TranscriptFileUri']}."
            )
        break
    else:
        print(f"Waiting for {response['TranscriptionJob']['TranscriptionJobName']}. Current status is {job_status}.")
    time.sleep(10)

The transcription job will take a few minutes to complete.

When the job’s complete, you can inspect the transcription output and check for the plain text transcript that was generated:

import json
# Get the Transcribe Output JSON file
s3 = boto3.client('s3')
output_bucket = job['TranscriptionJob']['Transcript']['TranscriptFileUri'].split('https://')[1].split('/',2)[1]
output_file_key = job['TranscriptionJob']['Transcript']['TranscriptFileUri'].split('https://')[1].split('/',2)[2]
s3_response_object = s3.get_object(Bucket=output_bucket, Key=output_file_key)
object_content = s3_response_object['Body'].read()

transcription_output = json.loads(object_content)

# Let's see what we have inside the job output JSON
print(transcription_output['results']['transcripts'][0]['transcript'])

Thank you for calling Big Jim’s Auto. This is Travis. How can I help you today? Hello, my name is Violet King and I bought a car not too long ago and a light is coming on um a light on the dashboard. And so I was wondering what I should do about that. Ok. It may depend on what kind of light we’re looking at here today, ma’am. Uh Could I get your first and last name spelled out for me so I can just get some information pulled up? Yes. My name is Violet Vviolet. My last name is King King. Ok, I got the call last week. Ok. Uh And what kind of car are we examining today? It’s, it’s a Ford Fusion. It’s 2017, 2017 Ford Fusion. OK. And for verification, ma’am. Do you happen to know the purchase date of the car? Yes, it, it, it was last Tuesday, August 10th. You say the 10th? Ok. And can you describe to me what kind of light we’re looking at? Y yes, it’s uh I, I think it’s a, an oil, an oil light, an oil light? Ok. Ok. And uh just for clarity on my end, ma’am. Um, uh, is this the first call you’ve made regarding this? Yes. Ok. And, uh, this might be kind of a silly question because I know you just got the car. But sometimes they make me ask a silly question about how many miles has the car been driven since you bought it? Oh. I’m not sure. Should I check? Uh, no, that’s ok. I’ll just, I’ll just put in that. We don’t know at this time. It’s ok. Um, so, um, uh, under the warranty we offer, um, we, the, we don’t handle in house oil changes. Um, we basically, when, when someone buys a car from us, the warranty, we have, it, it covers some stuff like, um, weather damage. Um, and, uh, if the engine light comes on, we take a look at that, but the oil change is something that we just don’t have, uh, here at the dealership that’s a little bit outsourced out and they’re pretty backed up right now because a lot of people have been, uh, staying in due to the recent pandemic and now everyone’s just starting to get out and a whole bunch of places are just completely bogged down. So we have a place that we typically outsource to, um, and they’re, they’re pretty reasonable. They’re about, I wanna say somewhere between 25 and $35 to do an oil change. So it’s really not that bad, but they’re a little bit backed up right now from what I’ve heard, I would recommend giving them a call as soon as you can before they close. Ok. What is their number? Uh, give me a second. Let me just rustle through the desk here. See if I can find their information. Uh, ok. Ok. Yes, I’m all right with them one moment, please. Sure. Ok. All their number is 888 333 2222. Ok, and they can fix my car. Yeah, they should be able to handle the oil change. I’m sorry, that’s not something that we cover under the warranty that uh, we have, um, but they should be able to get you settled and, uh, sorted. Ok. Ok. Thank you. Cool. No problem. Have a good one. Thank you. Yup. Bye.

After you have validated the existence of the text, you can use Amazon Bedrock to analyze the output:

# Now let's use this transcript to extract insights with a help of a Large Language Model on Amazon Bedrock
# First let's initialize the bedrock runtime client to invoke the model. 
bedrock_runtime = boto3.client('bedrock-runtime')
# Selecting Claude 3 Sonnet
model_id = 'anthropic.claude-3-sonnet-20240229-v1:0'

Using the transcription from the recorded call center call, we use Amazon Bedrock to call an FM (we use Anthropic’s Claude 3 Sonnet in this case), but you can choose from the other language models available on Amazon Bedrock.

You can now perform additional tasks.

Summarize the call between agent and client with Amazon Bedrock

The following prompt provides instructions to summarize the call discussion from the transcription:

#Summarize the call between agent and customer
summarize_prompt = """Based on the contents of <transcript></transcript>,summarize the call between agent and customer with focus on resolution <transcript></transcript> . 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": summarize_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
summarization = response_body['content'][0]['text']
print(summarization)

We get the following output:

Based on the contents of the transcript, here is a summary of the call between the agent (Travis) and the customer (Violet King) with a focus on resolution:

Violet King called about a light on the dashboard of her recently purchased 2017 Ford Fusion from Big Jim’s Auto. The light appeared to be an oil light. Travis explained that while their warranty covers certain issues like weather damage and engine lights, it does not cover oil changes. He recommended calling an outsourced oil change service that Big Jim’s Auto typically uses, which charges between $25-35 for an oil change.

Travis provided the phone number for the oil change service (888-333-2222) and mentioned that they are currently backed up due to the recent pandemic. He advised Violet to call them as soon as possible before they close to get her oil changed and resolve the issue with the oil light on her dashboard.

The resolution was for Violet to contact the recommended third-party oil change service to have her car’s oil changed, which should address the oil light issue she was experiencing with her newly purchased vehicle.

Extract the main topics with Amazon Bedrock

The following prompt provides instructions to extract the main topics discussed in the conversation from the transcription:

# Extracting the main topics from the conversation
maintopic_prompt = """The contents of <transcript></transcript> are a transcription of a conversation between agent and client. Based on the information, extract the main topics from the conversation  <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": maintopic_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
main_topic = response_body['content'][0]['text']
print(main_topic)

We get the following output:

Based on the conversation transcript, the main topics appear to be:

1. Dashboard warning light (specifically an oil light) on a recently purchased 2017 Ford Fusion.
2. Determining if the issue is covered under the warranty provided by the dealership (Big Jim’s Auto).
3. Recommendation to contact an external auto service provider (phone number provided) for an oil change service, as the dealership does not handle oil changes in-house.
4. Confirming that the external auto service provider can likely resolve the oil light issue by performing an oil change.

Extract the key phrases with Amazon Bedrock

The following prompt provides instructions to extract the key phrases discussed in the conversation from the transcription:

# Extracting the key phrases with Amazon Bedrock
keyphrase_prompt = """The contents of <transcript></transcript> are a transcription of a conversation between agent and client. Based on the information, extract the key phrases discussed in the conversation <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": keyphrase_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
keyphrases = response_body['content'][0]['text']
print(keyphrases)

We get the following output:

Based on the conversation transcript, here are the key phrases discussed:

• Oil light
• 2017 Ford Fusion
• Purchase date: August 10th
• First call regarding the issue
• Car mileage unknown
• Warranty does not cover oil changes
• Outsourced oil change service recommended
• Oil change service contact number: 888-333-2222
• Oil change service cost: $25 – $35
• Service is backed up due to the pandemic

Extract the reason why the client called the call center with Amazon Bedrock

The following prompt provides instructions to extract the reason for this client call to the call center from the transcription:

# Extracting the reason why client called the call center
reason_prompt = """The content of <transcript></transcript> is  transcription of a conversation between agent and client. Based on the information, extract the reason why client called the call center <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": reason_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
reason = response_body['content'][0]['text']
print(reason)

We get the following output:

Based on the transcription, the client called the call center because a light (specifically an oil light) was coming on the dashboard of their recently purchased 2017 Ford Fusion car. The client was seeking guidance on what to do about the oil light being on.

Extract the level of customer satisfaction with Amazon Bedrock

The following prompt provides instructions to extract the level of customer satisfaction experienced by the client from the transcription:

# Extracting the level of Customer Satisfaction 
satisfaction_prompt = """The content of <transcript></transcript> is  transcription of a conversation between agent and client. Based on the information, extract the level of customer satisfaction <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": satisfaction_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
csat= response_body['content'][0]['text']
print(csat)

We get the following output:

Based on the transcript, the level of customer satisfaction seems to be moderate.

Evidence:

1. The agent provided clear explanations regarding the issue with the oil light and why oil changes are not covered under their warranty.
2. The agent offered a recommendation for an external service provider that could perform the oil change, along with their contact information.
3. The customer acknowledged the information provided by the agent, indicating some level of satisfaction with the response.

However, there are no explicit statements from the customer expressing high satisfaction or dissatisfaction. The interaction remains polite and resolves the customer’s initial query, but there is no strong indication of exceptional satisfaction or disappointment.

Obtain the overall customer sentiment with Amazon Bedrock

The following prompt provides instructions to obtain the overall customer sentiment from the transcription:

# Extracting the overall customer sentiment.
sentiment_prompt = """The content of <transcript></transcript> is transcription of a conversation between agent and client. Based on the information, what is the overall customer sentiment of the conversation <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": sentiment_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
sentiment = response_body['content'][0]['text']
print(sentiment)

We get the following output:

Based on the conversation transcript, the overall customer sentiment seems neutral to slightly positive. Although the customer, Violet King, was initially concerned about a warning light on her recently purchased car’s dashboard, the agent (Travis) explained the situation clearly and provided a recommendation for getting an oil change from a third-party service provider. The customer acknowledged and accepted the suggestion without expressing significant frustration or dissatisfaction. The conversation ended on a polite note with the customer thanking the agent.

Obtain sentiment about the products or services discussed during the call with Amazon Bedrock

The following prompt provides instructions to obtain sentiment about products and services discussed during the call from the transcription:

## Obtaining sentiment about the products or services discussed.
sentiment_product_prompt = """The content of <transcript></transcript> is transcription of a conversation between agent and client. Based on the information, what is the overall sentiment about the products discussed in the conversation <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": sentiment_product_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
sentiment_product = response_body['content'][0]['text']
print(sentiment_product)

We get the following output:

Based on the conversation transcript, the overall sentiment about the products (the 2017 Ford Fusion car) discussed seems neutral to slightly positive. There are no major complaints or strong negative sentiments expressed about the car itself. The issue raised is regarding an oil light coming on, which seems to be a routine maintenance matter that needs to be addressed by getting an oil change. The agent helpfully provides the customer with information on where to get an oil change done, although mentioning that the warranty does not cover routine oil changes at the dealership. Overall, there is no indication of major dissatisfaction with the car itself from the conversation.

Create an email to send to the client with the call summary

The following prompt provides instructions to create an email to send to the client after the call is completed, summarizing the call:

## Creating an email to send to the client with the call summary
email_prompt = """The content of <transcript></transcript> is transcription of a conversation between agent and client. Based on the information, create an email to send to the client after the call is completed summarizing the call. 
User formal but friendly english style. <transcript></transcript>. 

<transcript>
{transcript}
</transcript>
"""

user_message =  {"role": "user", "content": email_prompt.format(transcript = transcription_output['results']['transcripts'][0]['transcript'])}

messages = [user_message]

body=json.dumps(
    {
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1000,
        "messages": messages
    }
)
bedrock_response = bedrock_runtime.invoke_model(body=body, modelId=model_id)
response_body = json.loads(bedrock_response.get('body').read())
email_conversation = response_body['content'][0]['text']
print(email_conversation)

We get the following output:

Dear Ms. Violet King,

I hope this email finds you well. I’m writing to summarize our conversation regarding the issue with your 2017 Ford Fusion that you purchased last Tuesday, August 10th.

During our call, you mentioned that an oil light was illuminated on the dashboard of your vehicle. As discussed, our dealership’s warranty does not cover routine maintenance services like oil changes. However, we have a trusted partner that we typically recommend for such services.

The recommended auto service provider is reachable at 888-333-2222. They are experienced in handling oil changes and should be able to assist you with your vehicle’s needs. Please note that they have been experiencing a high volume of requests due to the recent pandemic, so it’s advisable to call them as soon as possible to schedule an appointment.

Conclusion

Using generative AI through Amazon Bedrock to analyze audio transcripts generated by Amazon Transcribe unlocks valuable insights that would otherwise remain hidden within the audio data. By combining the powerful speech-to-text capabilities of Amazon Transcribe with the natural language understanding and generation capabilities of LLMs like those available through Amazon Bedrock, businesses can more efficiently extract key information, generate summaries, identify topics and sentiments, and create new content from their audio and video assets. This approach not only saves time and resources compared to manual transcription and analysis, but also opens up new opportunities for using existing content in innovative ways.

Whether it’s repurposing marketing materials, quickly capturing key points from meetings, or improving customer experience through call center analytics, the combination of Amazon Transcribe and large language models (LLMs) on Amazon Bedrock provides a powerful solution for unlocking the full potential of audio data.As these use cases have demonstrated, this technology can be applied across various domains, from content creation and SEO optimization to business intelligence and customer service. By staying at the forefront of these advancements, organizations can gain a competitive edge by effectively harnessing the wealth of information contained within their audio and video repositories, driving insights, and making more informed decisions.

About the Authors

Ana Maria Echeverri is an AI/ML Worldwide Service Specialist at AWS, focused on driving adoption of generative AI speech analytics use cases. She has worked in the data and AI industry for over 30 years, with over 10 years focused on helping organizations grow their AI maturity and capabilities for successful execution of AI strategies.

Vishesh Jha is a Senior Solutions Architect at AWS. His area of interest lies in generative AI, and he has helped customers and partners get started with NLP using AWS services such as Amazon Bedrock, Amazon Transcribe, and Amazon SageMaker. He is an avid soccer fan, and in his free time enjoys watching and playing the sport. He also loves cooking, gaming, and traveling with his family.

Bala Krishna Jakka is a Technical Account manager at AWS, with a passion for contact center and generative AI technologies. With extensive expertise in helping organizations use cutting-edge solutions, he thrives on staying ahead of the curve in this rapidly evolving field. When not immersed in the realms of AI and customer experience, he finds joy in the game of cricket, showcasing his skills on the pitch. A devoted family man, he cherishes the moments spent with his loved ones, creating lasting memories and finding balance amidst the demands of his professional pursuits

Read More

Fine-tuning is a powerful approach in natural language processing (NLP) and generative AI, allowing businesses to tailor pre-trained large language models (LLMs) for specific tasks. This process involves updating the model’s weights to improve its performance on targeted applications. By fine-tuning, the LLM can adapt its knowledge base to specific data and tasks, resulting in enhanced task-specific capabilities. To achieve optimal results, having a clean, high-quality dataset is of paramount importance. A well-curated dataset forms the foundation for successful fine-tuning. Additionally, careful adjustment of hyperparameters such as learning rate multiplier and batch size plays a crucial role in optimizing the model’s adaptation to the target task.

The capabilities in Amazon Bedrock for fine-tuning LLMs offer substantial benefits for enterprises. This feature enables companies to optimize models like Anthropic’s Claude 3 Haiku on Amazon Bedrock for custom use cases, potentially achieving performance levels comparable to or even surpassing more advanced models such as Anthropic’s Claude 3 Opus or Anthropic’s Claude 3.5 Sonnet. The result is a significant improvement in task-specific performance, while potentially reducing costs and latency. This approach offers a versatile solution to satisfy your goals for performance and response time, allowing businesses to balance capability, domain knowledge, and efficiency in your AI-powered applications.

In this post, we explore the best practices and lessons learned for fine-tuning Anthropic’s Claude 3 Haiku on Amazon Bedrock. We discuss the important components of fine-tuning, including use case definition, data preparation, model customization, and performance evaluation. This post dives deep into key aspects such as hyperparameter optimization, data cleaning techniques, and the effectiveness of fine-tuning compared to base models. We also provide insights on how to achieve optimal results for different dataset sizes and use cases, backed by experimental data and performance metrics.

As part of this post, we first introduce general best practices for fine-tuning Anthropic’s Claude 3 Haiku on Amazon Bedrock, and then present specific examples with the TAT- QA dataset (Tabular And Textual dataset for Question Answering).

Recommended use cases for fine-tuning

The use cases that are the most well-suited for fine-tuning Anthropic’s Claude 3 Haiku include the following:

• Classification – For example, when you have 10,000 labeled examples and want Anthropic’s Claude 3 Haiku to do well at this task.
• Structured outputs – For example, when you have 10,000 labeled examples specific to your use case and need Anthropic’s Claude 3 Haiku to accurately identify them.
• Tools and APIs – For example, when you need to teach Anthropic’s Claude 3 Haiku how to use your APIs well.
• Particular tone or language – For example, when you need Anthropic’s Claude 3 Haiku to respond with a particular tone or language specific to your brand.

Fine-tuning Anthropic’s Claude 3 Haiku has demonstrated superior performance compared to few-shot prompt engineering on base Anthropic’s Claude 3 Haiku, Anthropic’s Claude 3 Sonnet, and Anthropic’s Claude 3.5 Sonnet across various tasks. These tasks include summarization, classification, information retrieval, open-book Q&A, and custom language generation such as SQL. However, achieving optimal performance with fine-tuning requires effort and adherence to best practices.

To better illustrate the effectiveness of fine-tuning compared to other approaches, the following table provides a comprehensive overview of various problem types, examples, and their likelihood of success when using fine-tuning versus prompting with Retrieval Augmented Generation (RAG). This comparison can help you understand when and how to apply these different techniques effectively.

Prerequisites

Before diving into the best practices and optimizing fine-tuning LLMs on Amazon Bedrock, familiarize yourself with the general process and how-to outlined in Fine-tune Anthropic’s Claude 3 Haiku in Amazon Bedrock to boost model accuracy and quality. The post provides essential background information and context for the fine-tuning process, including step-by-step guidance on fine-tuning Anthropic’s Claude 3 Haiku on Amazon Bedrock both through the Amazon Bedrock console and Amazon Bedrock API.

LLM fine-tuning lifecycle

The process of fine-tuning an LLM like Anthropic’s Claude 3 Haiku on Amazon Bedrock typically follows these key stages:

• Use case definition – Clearly define the specific task or knowledge domain for fine-tuning
• Data preparation – Gather and clean high-quality datasets relevant to the use case
• Data formatting – Structure the data following best practices, including semantic blocks and system prompts where appropriate
• Model customization – Configure the fine-tuning job on Amazon Bedrock, setting parameters like learning rate and batch size, enabling features like early stopping to prevent overfitting
• Training and monitoring – Run the training job and monitor the status of training job
• Performance evaluation – Assess the fine-tuned model’s performance against relevant metrics, comparing it to base models
• Iteration and deployment – Based on the result, refine the process if needed, then deploy the model for production

Throughout this journey, depending on the business case, you may choose to combine fine-tuning with techniques like prompt engineering for optimal results. The process is inherently iterative, allowing for continuous improvement as new data or requirements emerge.

Use case and dataset

The TAT-QA dataset is related to a use case for question answering on a hybrid of tabular and textual content in finance where tabular data is organized in table formats such as HTML, JSON, Markdown, and LaTeX. We focus on the task of answering questions about the table. The evaluation metric is the F1 score that measures the word-to-word matching of the extracted content between the generated output and the ground truth answer. The TAT-QA dataset has been divided into train (28,832 rows), dev (3,632 rows), and test (3,572 rows).

The following screenshot provides a snapshot of the TAT-QA data, which comprises a table with tabular and textual financial data. Following this financial data table, a detailed question-answer set is presented to demonstrate the complexity and depth of analysis possible with the TAT-QA dataset. This comprehensive table is from the paper TAT-QA: A Question Answering Benchmark on a Hybrid of Tabular and Textual Content in Finance, and it includes several key components:

• Reasoning types – Each question is categorized by the type of reasoning required
• Questions – A variety of questions that test different aspects of understanding and interpreting the financial data
• Answers – The correct responses to each question, showcasing the precision required in financial analysis
• Scale – Where applicable, the unit of measurement for the answer
• Derivation – For some questions, the calculation or logic used to arrive at the answer is provided

The following screenshot shows a formatted version of the data as JSONL and is passed to Anthropic’s Claude 3 Haiku for fine-tuning training data. The preceding table has been structured in JSONL format with system, user role (which contains the data and the question), and assistant role (which has answers). The table is enclosed within the XML tag <table><table>, helping Anthropic’s Claude 3 Haiku parse the prompt with the data from the table. For the model fine-tuning and performance evaluation, we randomly selected 10,000 examples from the TAT-QA dataset to fine-tune the model, and randomly picked 3,572 records from the remainder of the dataset as testing data.

Best practices for data cleaning and data validation

When fine-tuning the Anthropic’s Claude 3 Haiku model, the quality of training data is paramount and serves as the primary determinant of the output quality, surpassing the importance of any other step in the fine-tuning process. Our experiments have consistently shown that high-quality datasets, even if smaller in size, yield better results than a larger but less refined one. This “quality over quantity” approach should guide the entire data preparation process. Data cleaning and validation are essential steps in maintaining the quality of the training set. The following are two effective methods:

• Human evaluation – This method involves subject matter experts (SMEs) manually reviewing each data point for quality and relevance. Though time-consuming, it provides unparalleled insight into the nuances of the specific tasks.
• LLM as a judge – For large datasets, using Anthropic’s Claude models as a judge can be more efficient. For example, you can use Anthropic’s Claude 3.5 Sonnet as a judge to decide whether each provided training record meets the high quality requirement. The following is an example prompt template:

{'prompt': {
'system': "You are a reliable and impartial expert judge in question/answering data assessment. ",
'messages': [
{'role': 'user', 'content': [{'type': 'text', 'text': 'Your task is to take a question, an answer, and a context which may include multiple documents, and provide a judgment on whether the answer to the question is correct or not. This decision should be based either on the provided context or your general knowledge and memory. If the answer contradicts the information in context, it's incorrect. A correct answer is ideally derived from the given context. If no context is given, a correct answer should be factually true and directly and unambiguously address the question.nnProvide a short step-by-step reasoning with a maximum of 4 sentences within the <reason></reason> xml tags and provide a single correct or incorrect response within the <judgement></judgement> xml tags.n <context>n...n</context>n<question>n...n</question>n<answer>n...n</answer>n'}]}]}}

The following is a sample output from Anthropic’s Claude 3.5 Sonnet:

{'id': 'job_id',
 'type': 'message',
 'role': 'assistant',
 'model': 'claude-3-5-sonnet-20240620',
 'content': [{'type': 'text',
   'text': '<reason>n1. I'll check the table for information... </reason>nn<judgement>correct</judgement>'}],
 'stop_reason': 'end_turn',
 'stop_sequence': None,
 'usage': {'input_tokens': 923, 'output_tokens': 90}}

This LLM-as-a-judge approach is effective for large datasets, allowing for efficient and consistent quality assessment across a wide range of examples. It can help identify and filter out low-quality or irrelevant data points, making sure only the most suitable examples are used for fine-tuning.

The format of your training data is equally important. Although it’s optional, it’s highly recommended to include a system prompt that clearly defines the model’s role and tasks. In addition, including rationales within XML tags can provide valuable context for the model and facilitate extraction of key information. Prompt optimization is one of the key factors in improving model performance. Following established guidelines, such as those provided by Anthropic, can significantly enhance results. This might include structuring prompts with semantic blocks within XML tags, both in training samples and at inference time.

By adhering to these best practices in data cleaning, validation, and formatting, you can create a high-quality dataset that forms the foundation for successful fine-tuning. In the world of model training, quality outweighs quantity, and a well-prepared dataset is key to unlocking the full potential of fine-tuning Anthropic’s Claude 3 Haiku.

Best practices for performing model customization training jobs

When fine-tuning Anthropic’s Claude 3 Haiku on Amazon Bedrock, it’s crucial to optimize your training parameters to achieve the best possible performance. Our experiments have revealed several key insights that can guide you in effectively setting up your customization training jobs.

One of the most critical aspects of fine-tuning is selecting the right hyperparameters, particularly learning rate multiplier and batch size (see the appendix in this post for definitions). Our experiment results have shown that these two factors can significantly impact the model’s performance, with improvements ranging from 2–10% across different tasks. For the learning rate multiplier, the value ranges between 0.1–2.0, with a default value of 1.0. We suggest starting with the default value and potentially adjusting this value based on your evaluation result. Batch size is another important parameter, and its optimal value can vary depending on your dataset size. Based on our hyperparameter tuning experiments across different use cases, the API allows a range of 4–256, with a default of 32. However, we’ve observed that dynamically adjusting the batch size based on your dataset size can lead to better results:

• For datasets with 1,000 or more examples, aim for a batch size between 32–64
• For datasets between 500–1,000 examples, a batch size between 16–32 is generally suitable
• For smaller datasets with fewer than 500 examples, consider a batch size between 4–16

The following chart illustrates how model performance improves as the size of the training dataset increases, as well as the change of optimal parameters, using the TAT-QA dataset. Each data point is annotated with the optimal learning rate multiplier (LRM), batch size (BS), and number of epochs (Epoch) used to achieve the best performance with the dataset size. We can observe that larger datasets tend to benefit from higher learning rates and batch sizes, whereas smaller datasets require more training epochs. The red dashed line is the baseline Anthropic’s Claude 3 Haiku performance without fine-tuning efforts.

By following these guidelines, you can configure an Anthropic’s Claude 3 Haiku fine-tuning job with a higher chance of success. However, remember that these are general recommendations and the optimal settings may vary depending on your specific use case and dataset characteristics.

In scenarios with large amounts of data (1,000–10,000 examples), the learning rate tends to have a more significant impact on performance. Conversely, for smaller datasets (32–100 examples), the batch size becomes the dominant factor.

Performance evaluations

The fine-tuned Anthropic’s Claude 3 Haiku model demonstrated substantial performance improvements over base models when evaluated on the financial Q&A task, highlighting the effectiveness of the fine-tuning process on specialized data. Based on the evaluation results, we found the following:

• Fine-tuned Anthropic’s Claude 3 Haiku performed better than Anthropic’s Claude 3 Haiku, Anthropic’s Claude 3 Sonnet, and Anthropic’s Claude 3.5 Sonnet for TAT-QA dataset across the target use case of question answering on financial text and tabular content.
• For the performance evaluation metric F1 score (see the appendix for definition), fine-tuned Anthropic’s Claude 3 Haiku achieved a score of 91.2%, which is a 24.60% improvement over the Anthropic’s Claude 3 Haiku base model’s score of 73.2%. Fine-tuned Anthropic’s Claude 3 Haiku also achieved a 19.6% improvement over the Anthropic’s Claude 3 Sonnet base model’s performance, which obtained an F1 score of 76.3%. Fine-tuned Anthropic’s Claude 3 Haiku even achieved better performance over the Anthropic’s Claude 3.5 Sonnet base model.

The following table provides a detailed comparison of the performance metrics for the fine-tuned Claude 3 Haiku model against various base models, illustrating the significant improvements achieved through fine-tuning.

Few-shot examples improve performance not only on the base model, but also on fine-tuned models, especially when the fine-tuning data is small.

Fine-tuning also demonstrated significant benefits in reducing token usage. On the TAT-QA HTML test set (893 examples), the fine-tuned Anthropic’s Claude 3 Haiku model reduced the average output token count by 35% compared to the base model, as shown in the following table.

We use the following figures to illustrate the token count distribution for both the base Anthropic’s Claude 3 Haiku and fine-tuned Anthropic’s Claude 3 Haiku models. The left graph shows the distribution for the base model, and the right graph displays the distribution for the fine-tuned model. These histograms demonstrate a shift towards more concise output in the fine-tuned model, with a notable reduction in the frequency of longer token sequences.

To further illustrate this improvement, consider the following example from the test set:

• Question: "How did the company adopt Topic 606?"
• Ground truth answer: "the modified retrospective method"
• Base Anthropic’s Claude 3 Haiku response: "The company adopted the provisions of Topic 606 in fiscal 2019 utilizing the modified retrospective method"
• Fine-tuned Anthropic’s Claude 3 Haiku response: "the modified retrospective method"

As evident from this example, the fine-tuned model produces a more concise and precise answer, matching the ground truth exactly, whereas the base model includes additional, unnecessary information. This reduction in token usage, combined with improved accuracy, can lead to enhanced efficiency and reduced costs in production deployments.

Conclusion

Fine-tuning Anthropic’s Claude 3 Haiku on Amazon Bedrock offers significant performance improvements for specialized tasks. Our experiments demonstrate that careful attention to data quality, hyperparameter optimization, and best practices in the fine-tuning process can yield substantial gains over base models. Key takeaways include the following:

• The importance of high-quality, task-specific datasets, even if smaller in size
• Optimal hyperparameter settings vary based on dataset size and task complexity
• Fine-tuned models consistently outperform base models across various metrics
• The process is iterative, allowing for continuous improvement as new data or requirements emerge

Although fine-tuning provides impressive results, combining it with other techniques like prompt engineering may lead to even better outcomes. As LLM technology continues to evolve, mastering fine-tuning techniques will be crucial for organizations looking to use these powerful models for specific use cases and tasks.

Now you’re ready to fine-tune Anthropic’s Claude 3 Haiku on Amazon Bedrock for your use case. We look forward to seeing what you build when you put this new technology to work for your business.

Appendix

We used the following hyperparameters as part of our fine-tuning:

• Learning rate multiplier – Learning rate multiplier is one of the most critical hyperparameters in LLM fine-tuning. It influences the learning rate at which model parameters are updated after each batch.
• Batch size – Batch size is the number of training examples processed in one iteration. It directly impacts GPU memory consumption and training dynamics.
• Epoch – One epoch means the model has seen every example in the dataset one time. The number of epochs is a crucial hyperparameter that affects model performance and training efficiency.

For our evaluation, we used the F1 score, which is an evaluation metric to assess the performance of LLMs and traditional ML models.

To compute the F1 score for LLM evaluation, we need to define precision and recall at the token level. Precision measures the proportion of generated tokens that match the reference tokens, and recall measures the proportion of reference tokens that are captured by the generated tokens. The F1 score ranges from 0–100, with 100 being the best possible score and 0 being the lowest. However, interpretation can vary depending on the specific task and requirements.

We calculate these metrics as follows:

• Precision = (Number of matching tokens in generated text) / (Total number of tokens in generated text)
• Recall = (Number of matching tokens in generated text) / (Total number of tokens in reference text)
• F1 = (2 * (Precision * Recall) / (Precision + Recall)) * 100

For example, let’s say the LLM generates the sentence “The cat sits on the mat in the sun” and the reference sentence is “The cat sits on the soft mat under the warm sun.” The precision would be 6/9 (6 matching tokens out of 9 generated tokens), and the recall would be 6/11 (6 matching tokens out of 11 reference tokens).

• Precision = 6/9 ≈ 0.667
• Recall = 6/11 ≈ 0.545
• F1 score = (2 * (0.667 * 0.545) / (0.667 + 0.545)) * 100 ≈ 59.90

About the Authors

Yanyan Zhang is a Senior Generative AI Data Scientist at Amazon Web Services, where she has been working on cutting-edge AI/ML technologies as a Generative AI Specialist, helping customers use generative AI to achieve their desired outcomes. Yanyan graduated from Texas A&M University with a PhD in Electrical Engineering. Outside of work, she loves traveling, working out, and exploring new things.

Sovik Kumar Nath is an AI/ML and Generative AI Senior Solutions Architect with AWS. He has extensive experience designing end-to-end machine learning and business analytics solutions in finance, operations, marketing, healthcare, supply chain management, and IoT. He has double master’s degrees from the University of South Florida and University of Fribourg, Switzerland, and a bachelor’s degree from the Indian Institute of Technology, Kharagpur. Outside of work, Sovik enjoys traveling, and adventures.

Jennifer Zhu is a Senior Applied Scientist at AWS Bedrock, where she helps building and scaling generative AI applications with foundation models. Jennifer holds a PhD degree from Cornell University, and a master degree from University of San Francisco. Outside of work, she enjoys reading books and watching tennis games.

Fang Liu is a principal machine learning engineer at Amazon Web Services, where he has extensive experience in building AI/ML products using cutting-edge technologies. He has worked on notable projects such as Amazon Transcribe and Amazon Bedrock. Fang Liu holds a master’s degree in computer science from Tsinghua University.

Yanjun Qi is a Senior Applied Science Manager at the Amazon Bedrock Science. She innovates and applies machine learning to help AWS customers speed up their AI and cloud adoption.

Read More

As enterprises increasingly embrace generative AI , they face challenges in managing the associated costs. With demand for generative AI applications surging across projects and multiple lines of business, accurately allocating and tracking spend becomes more complex. Organizations need to prioritize their generative AI spending based on business impact and criticality while maintaining cost transparency across customer and user segments. This visibility is essential for setting accurate pricing for generative AI offerings, implementing chargebacks, and establishing usage-based billing models.

Without a scalable approach to controlling costs, organizations risk unbudgeted usage and cost overruns. Manual spend monitoring and periodic usage limit adjustments are inefficient and prone to human error, leading to potential overspending. Although tagging is supported on a variety of Amazon Bedrock resources—including provisioned models, custom models, agents and agent aliases, model evaluations, prompts, prompt flows, knowledge bases, batch inference jobs, custom model jobs, and model duplication jobs—there was previously no capability for tagging on-demand foundation models. This limitation has added complexity to cost management for generative AI initiatives.

To address these challenges, Amazon Bedrock has launched a capability that organization can use to tag on-demand models and monitor associated costs. Organizations can now label all Amazon Bedrock models with AWS cost allocation tags, aligning usage to specific organizational taxonomies such as cost centers, business units, and applications. To manage their generative AI spend judiciously, organizations can use services like AWS Budgets to set tag-based budgets and alarms to monitor usage, and receive alerts for anomalies or predefined thresholds. This scalable, programmatic approach eliminates inefficient manual processes, reduces the risk of excess spending, and ensures that critical applications receive priority. Enhanced visibility and control over AI-related expenses enables organizations to maximize their generative AI investments and foster innovation.

Introducing Amazon Bedrock application inference profiles

Amazon Bedrock recently introduced cross-region inference, enabling automatic routing of inference requests across AWS Regions. This feature uses system-defined inference profiles (predefined by Amazon Bedrock), which configure different model Amazon Resource Names (ARNs) from various Regions and unify them under a single model identifier (both model ID and ARN). While this enhances flexibility in model usage, it doesn’t support attaching custom tags for tracking, managing, and controlling costs across workloads and tenants.

To bridge this gap, Amazon Bedrock now introduces application inference profiles, a new capability that allows organizations to apply custom cost allocation tags to track, manage, and control their Amazon Bedrock on-demand model costs and usage. This capability enables organizations to create custom inference profiles for Bedrock base foundation models, adding metadata specific to tenants, thereby streamlining resource allocation and cost monitoring across varied AI applications.

Creating application inference profiles

Application inference profiles allow users to define customized settings for inference requests and resource management. These profiles can be created in two ways:

1. Single model ARN configuration: Directly create an application inference profile using a single on-demand base model ARN, allowing quick setup with a chosen model.
2. Copy from system-defined inference profile: Copy an existing system-defined inference profile to create an application inference profile, which will inherit configurations such as cross-Region inference capabilities for enhanced scalability and resilience.

The application inference profile ARN has the following format, where the inference profile ID component is a unique 12-digit alphanumeric string generated by Amazon Bedrock upon profile creation.

arn:aws:bedrock:<region>:<account_id>:application-inference-profile/<inference_profile_id>

System-defined compared to application inference profiles

The primary distinction between system-defined and application inference profiles lies in their type attribute and resource specifications within the ARN namespace:

• System-defined inference profiles: These have a type attribute of SYSTEM_DEFINED and utilize the inference-profile resource type. They’re designed to support cross-Region and multi-model capabilities but are managed centrally by AWS.

  {
   …
  "inferenceProfileArn": "arn:aws:bedrock:us-east-1:<Account ID>:inference-profile/us-1.anthropic.claude-3-sonnet-20240229-v1:0",
  "inferenceProfileId": "us-1.anthropic.claude-3-sonnet-20240229-v1:0",
  "inferenceProfileName": "US-1 Anthropic Claude 3 Sonnet",
  "status": "ACTIVE",
  "type": "SYSTEM_DEFINED",
  …
  }

• Application inference profiles: These profiles have a type attribute of APPLICATION and use the application-inference-profile resource type. They’re user-defined, providing granular control and flexibility over model configurations and allowing organizations to tailor policies with attribute-based access control (ABAC) using AWS Identity and Access Management (IAM). This enables more precise IAM policy authoring to manage Amazon Bedrock access more securely and efficiently.

  {
  …
  "inferenceProfileArn": "arn:aws:bedrock:us-east-1:<Account ID>:application-inference-profile/<Auto generated ID>",
  "inferenceProfileId": <Auto generated ID>,
  "inferenceProfileName": <User defined name>,
  "status": "ACTIVE",
  "type": "APPLICATION"
  …
  }

These differences are important when integrating with Amazon API Gateway or other API clients to help ensure correct model invocation, resource allocation, and workload prioritization. Organizations can apply customized policies based on profile type, enhancing control and security for distributed AI workloads. Both models are shown in the following figure.

Establishing application inference profiles for cost management

Imagine an insurance provider embarking on a journey to enhance customer experience through generative AI. The company identifies opportunities to automate claims processing, provide personalized policy recommendations, and improve risk assessment for clients across various regions. However, to realize this vision, the organization must adopt a robust framework for effectively managing their generative AI workloads.

The journey begins with the insurance provider creating application inference profiles that are tailored to their diverse business units. By assigning AWS cost allocation tags, the organization can effectively monitor and track their Bedrock spend patterns. For example, the claims processing team established an application inference profile with tags such as dept:claims, team:automation, and app:claims_chatbot. This tagging structure categorizes costs and allows assessment of usage against budgets.

Users can manage and use application inference profiles using Bedrock APIs or the boto3 SDK:

• CreateInferenceProfile: Initiates a new inference profile, allowing users to configure the parameters for the profile.
• GetInferenceProfile: Retrieves the details of a specific inference profile, including its configuration and current status.
• ListInferenceProfiles: Lists all available inference profiles within the user’s account, providing an overview of the profiles that have been created.
• TagResource: Allows users to attach tags to specific Bedrock resources, including application inference profiles, for better organization and cost tracking.
• ListTagsForResource: Fetches the tags associated with a specific Bedrock resource, helping users understand how their resources are categorized.
• UntagResource: Removes specified tags from a resource, allowing for management of resource organization.
• Invoke models with application inference profiles:

• • Converse API: Invokes the model using a specified inference profile for conversational interactions.
  • ConverseStream API: Similar to the Converse API but supports streaming responses for real-time interactions.
  • InvokeModel API: Invokes the model with a specified inference profile for general use cases.
  • InvokeModelWithResponseStream API: Invokes the model and streams the response, useful for handling large data outputs or long-running processes.

Note that application inference profile APIs cannot be accessed through the AWS Management Console.

Invoke model with application inference profile using Converse API

The following example demonstrates how to create an application inference profile and then invoke the Converse API to engage in a conversation using that profile –

def create_inference_profile(profile_name, model_arn, tags):
    """Create Inference Profile using base model ARN"""
    response = bedrock.create_inference_profile(
        inferenceProfileName=profile_name,
        description="test",
        modelSource={'copyFrom': model_arn},
        tags=tags
    )
    print("CreateInferenceProfile Response:", response['ResponseMetadata']['HTTPStatusCode']),
    print(f"{response}n")
    return response

# Create Inference Profile
print("Testing CreateInferenceProfile...")
tags = [{'key': 'dept', 'value': 'claims'}]
base_model_arn = "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-3-sonnet-20240229-v1:0"
claims_dept_claude_3_sonnet_profile = create_inference_profile("claims_dept_claude_3_sonnet_profile", base_model_arn, tags)

# Extracting the ARN and retrieving Application Inference Profile ID
claims_dept_claude_3_sonnet_profile_arn = claims_dept_claude_3_sonnet_profile['inferenceProfileArn']

def converse(model_id, messages):
    """Use the Converse API to engage in a conversation with the specified model"""
    response = bedrock_runtime.converse(
        modelId=model_id,
        messages=messages,
        inferenceConfig={
            'maxTokens': 300,  # Specify max tokens if needed
        }
    )
    
    status_code = response.get('ResponseMetadata', {}).get('HTTPStatusCode')
    print("Converse Response:", status_code)
    parsed_response = parse_converse_response(response)
    print(parsed_response)
    return response

# Example of Converse API with Application Inference Profile
print("nTesting Converse...")
prompt = "nnHuman: Tell me about Amazon Bedrock.nnAssistant:"
messages = [{"role": "user", "content": [{"text": prompt}]}]
response = converse(claims_dept_claude_3_sonnet_profile_arn, messages)

Tagging, resource management, and cost management with application inference profiles

Tagging within application inference profiles allows organizations to allocate costs with specific generative AI initiatives, ensuring precise expense tracking. Application inference profiles enable organizations to apply cost allocation tags at creation and support additional tagging through the existing TagResource and UnTagResource APIs, which allow metadata association with various AWS resources. Custom tags such as project_id, cost_center, model_version, and environment help categorize resources, improving cost transparency and allowing teams to monitor spend and usage against budgets.

Visualize cost and usage with application inference profiles and cost allocation tags

Leveraging cost allocation tags with tools like AWS Budgets, AWS Cost Anomaly Detection, AWS Cost Explorer, AWS Cost and Usage Reports (CUR), and Amazon CloudWatch provides organizations insights into spending trends, helping detect and address cost spikes early to stay within budget.

With AWS Budgets, organization can set tag-based thresholds and receive alerts as spending approach budget limits, offering a proactive approach to maintaining control over AI resource costs and quickly addressing any unexpected surges. For example, a $10,000 per month budget could be applied on a specific chatbot application for the Support Team in the Sales Department by applying the following tags to the application inference profile: dept:sales, team:support, and app:chat_app. AWS Cost Anomaly Detection can also monitor tagged resources for unusual spending patterns, making it easier to operationalize cost allocation tags by automatically identifying and flagging irregular costs.

The following AWS Budgets console screenshot illustrates an exceeded budget threshold:

For deeper analysis, AWS Cost Explorer and CUR enable organizations to analyze tagged resources daily, weekly, and monthly, supporting informed decisions on resource allocation and cost optimization. By visualizing cost and usage based on metadata attributes, such as tag key/value and ARN, organizations gain an actionable, granular view of their spending.

The following AWS Cost Explorer console screenshot illustrates a cost and usage graph filtered by tag key and value:

The following AWS Cost Explorer console screenshot illustrates a cost and usage graph filtered by Bedrock application inference profile ARN:

Organizations can also use Amazon CloudWatch to monitor runtime metrics for Bedrock applications, providing additional insights into performance and cost management. Metrics can be graphed by application inference profile, and teams can set alarms based on thresholds for tagged resources. Notifications and automated responses triggered by these alarms enable real-time management of cost and resource usage, preventing budget overruns and maintaining financial stability for generate AI workloads.

The following Amazon CloudWatch console screenshot highlights Bedrock runtime metrics filtered by Bedrock application inference profile ARN:

The following Amazon CloudWatch console screenshot highlights an invocation limit alarm filtered by Bedrock application inference profile ARN:

Through the combined use of tagging, budgeting, anomaly detection, and detailed cost analysis, organizations can effectively manage their AI investments. By leveraging these AWS tools, teams can maintain a clear view of spending patterns, enabling more informed decision-making and maximizing the value of their generative AI initiatives while ensuring critical applications remain within budget.

Retrieving application inference profile ARN based on the tags for Model invocation

Organizations often use a generative AI gateway or large language model proxy when calling Amazon Bedrock APIs, including model inference calls. With the introduction of application inference profiles, organizations need to retrieve the inference profile ARN to invoke model inference for on-demand foundation models. There are two primary approaches to obtain the appropriate inference profile ARN.

• Static configuration approach: This method involves maintaining a static configuration file in the AWS Systems Manager Parameter Store or AWS Secrets Manager that maps tenant/workload keys to their corresponding application inference profile ARNs. While this approach offers simplicity in implementation, it has significant limitations. As the number of inference profiles scales from tens to hundreds or even thousands, managing and updating this configuration file becomes increasingly cumbersome. The static nature of this method requires manual updates whenever changes occur, which can lead to inconsistencies and increased maintenance overhead, especially in large-scale deployments where organizations need to dynamically retrieve the correct inference profile based on tags.
• Dynamic retrieval using the Resource Groups API: The second, more robust approach leverages the AWS Resource Groups GetResources API to dynamically retrieve application inference profile ARNs based on resource and tag filters. This method allows for flexible querying using various tag keys such as tenant ID, project ID, department ID, workload ID, model ID, and region. The primary advantage of this approach is its scalability and dynamic nature, enabling real-time retrieval of application inference profile ARNs based on current tag configurations.

However, there are considerations to keep in mind. The GetResources API has throttling limits, necessitating the implementation of a caching mechanism. Organizations should maintain a cache with a Time-To-Live (TTL) based on the API’s output to optimize performance and reduce API calls. Additionally, implementing thread safety is crucial to help ensure that organizations always read the most up-to-date inference profile ARNs when the cache is being refreshed based on the TTL.

As illustrated in the following diagram, this dynamic approach involves a client making a request to the Resource Groups service with specific resource type and tag filters. The service returns the corresponding application inference profile ARN, which is then cached for a set period. The client can then use this ARN to invoke the Bedrock model through the InvokeModel or Converse API.

By adopting this dynamic retrieval method, organizations can create a more flexible and scalable system for managing application inference profiles, allowing for more straightforward adaptation to changing requirements and growth in the number of profiles.

The architecture in the preceding figure illustrates two methods for dynamically retrieving inference profile ARNs based on tags. Let’s describe both approaches with their pros and cons:

1. Bedrock client maintaining the cache with TTL: This method involves the client directly querying the AWS ResourceGroups service using the GetResources API based on resource type and tag filters. The client caches the retrieved keys in a client-maintained cache with a TTL. The client is responsible for refreshing the cache by calling the GetResources API in the thread safe way.
2. Lambda-based Method: This approach uses AWS Lambda as an intermediary between the calling client and the ResourceGroups API. This method employs Lambda Extensions core with an in-memory cache, potentially reducing the number of API calls to ResourceGroups. It also interacts with Parameter Store, which can be used for configuration management or storing cached data persistently.

Both methods use similar filtering criteria (resource-type-filter and tag-filters) to query the ResourceGroup API, allowing for precise retrieval of inference profile ARNs based on attributes such as tenant, model, and Region. The choice between these methods depends on factors such as the expected request volume, desired latency, cost considerations, and the need for additional processing or security measures. The Lambda-based approach offers more flexibility and optimization potential, while the direct API method is simpler to implement and maintain.

Overview of Amazon Bedrock resources tagging capabilities

The tagging capabilities of Amazon Bedrock have evolved significantly, providing a comprehensive framework for resource management across multi-account AWS Control Tower setups. This evolution enables organizations to manage resources across development, staging, and production environments, helping organizations track, manage, and allocate costs for their AI/ML workloads.

At its core, the Amazon Bedrock resource tagging system spans multiple operational components. Organizations can effectively tag their batch inference jobs, agents, custom model jobs, knowledge bases, prompts, and prompt flows. This foundational level of tagging supports granular control over operational resources, enabling precise tracking and management of different workload components. The model management aspect of Amazon Bedrock introduces another layer of tagging capabilities, encompassing both custom and base models, and distinguishes between provisioned and on-demand models, each with its own tagging requirements and capabilities.

With the introduction of application inference profiles, organizations can now manage and track their on-demand Bedrock base foundation models. Because teams can create application inference profiles derived from system-defined inference profiles, they can configure more precise resource tracking and cost allocation at the application level. This capability is particularly valuable for organizations that are running multiple AI applications across different environments, because it provides clear visibility into resource usage and costs at a granular level.

The following diagram visualizes the multi-account structure and demonstrates how these tagging capabilities can be implemented across different AWS accounts.

Conclusion

In this post we introduced the latest feature from Amazon Bedrock, application inference profiles. We explored how it operates and discussed key considerations. The code sample for this feature is available in this GitHub repository. This new capability enables organizations to tag, allocate, and track on-demand model inference workloads and spending across their operations. Organizations can label all Amazon Bedrock models using tags and monitoring usage according to their specific organizational taxonomy—such as tenants, workloads, cost centers, business units, teams, and applications. This feature is now generally available in all AWS Regions where Amazon Bedrock is offered.

About the authors

Kyle T. Blocksom is a Sr. Solutions Architect with AWS based in Southern California. Kyle’s passion is to bring people together and leverage technology to deliver solutions that customers love. Outside of work, he enjoys surfing, eating, wrestling with his dog, and spoiling his niece and nephew.

Dhawal Patel is a Principal Machine Learning Architect at AWS. He has worked with organizations ranging from large enterprises to mid-sized startups on problems related to distributed computing, and Artificial Intelligence. He focuses on Deep learning including NLP and Computer Vision domains. He helps customers achieve high performance model inference on SageMaker.

Read More

Traditionally, clinical trials not only place a significant burden on patients and participants due to the costs associated with transportation, lodging, meals, and dependent care, but also have an environmental impact. With the advancement of available technologies, decentralized clinical trials have become a widely popular topic of discussion and offer a more sustainable approach. Decentralized clinical trials reduce the need to travel to study sites by lowering the financial burden on all parties involved, thereby accelerating patient recruitment and reducing dropout rates. Decentralized clinical trials use technologies such as wearable devices, patient apps, smartphones, and telemedicine to accelerate recruitment, reduce dropout, and minimize the carbon footprint of clinical research. AWS can play a key role in enabling fast implementation of these decentralized clinical trials.

In this post, we discuss how to use AWS to support a decentralized clinical trial across the four main pillars of a decentralized clinical trial (virtual trials, personalized patient engagement, patient-centric trial design, and centralized data management). By exploring these AWS powered alternatives, we aim to demonstrate how organizations can drive progress towards more environmentally friendly clinical research practices.

The challenge and impact of sustainability on clinical trials

With the rise of greenhouse gas emissions globally, finding ways to become more sustainable is quickly becoming a challenge across all industries. At the same time, global health awareness and investments in clinical research have increased as a result of motivations by major events like the COVID-19 pandemic. For instance, in 2021, we saw a significant increase in awareness of clinical research studies seeking volunteers, which was reported at 63% compared to 54% in 2019 by Applied Clinical Trials. This suggests that the COVID-19 pandemic brought increased attention to clinical trials among the public and magnified the importance of including diverse populations in clinical research.

These clinical research trials study new tests and treatments while evaluating their effects on human health outcomes. People often volunteer to take part in clinical trials to test medical interventions, including drugs, biological products, surgical procedures, radiological procedures, devices, behavioral treatments, and preventive care. The rise of clinical trials presents a major sustainability challenge—they are often not sustainable and can contribute substantially to greenhouse gas emissions due to how they are being implemented. The main sources of these are usually associated with the intensive energy use associated with research premises and air travel.

This post discusses an alternative to clinical trials—by decentralizing clinical trials, we can reduce the major greenhouse gas emissions caused by human activities present in clinical trials today.

The CRASH trial case study

We can further examine the impact of carbon emissions associated with clinical trials through the carbon audit of the CRASH trial case lead by medical research journal, BMJ. The CRASH trial was a clinical trial conducted from 1999–2004 and recruited patients from 49 countries in the span of 5 years. In the study, the effect of intravenous corticosteroids (a drug produced by Pfizer) on death within 14 days in 10,008 adults with clinically significant head injuries was examined. BMJ conducted an audit on the total emissions of greenhouse gases that were produced by the trials and calculated that roughly 126 metric tons (carbon dioxide equivalent) was emitted during a 1-year period. Over a 5-year period, it would mean that the entire trial would be responsible for about 630 metric tons of carbon dioxide equivalent.

Much of these greenhouse gas emissions can be attributed to travel (such as air travel, hotel, meetings), distribution associated for drugs and documents, and electricity used in coordination centers. According to the EPA, the average passenger vehicle emits about 4.6 metric tons of carbon dioxide per year. In comparison, 630 tons of carbon dioxide would be equivalent to the annual emissions of around 137 passenger vehicles. Similarly, the average US household generates about 20 metric tons of carbon dioxide per year from energy use. 630 tons of carbon dioxide would also be equal to the annual emissions of around 31 average US homes. 630 tons of carbon dioxide already represents a very substantial amount of greenhouse gas for one clinical trial. According to sources from government databases and research institutions, there are around 300,000–600,000 clinical trials conducted globally each year, amplifying this impact by several hundred thousand times.

Clinical trials vs. decentralized clinical trials

Decentralized clinical trials present opportunities to address the sustainability challenges associated with traditional clinical trial models. As a byproduct of decentralized trials, there are also improvements in the patient experience by reducing their burden, making the process more convenient and sustainable.

Today, clinical trials can contribute significantly to greenhouse gas emissions, primarily through energy use in research facilities and air travel. In contrast to the energy-intensive nature of centralized trial sites, the distributed nature of decentralized clinical trials offers a more practical and cost-effective approach to implementing renewable energy solutions.

For centralized clinical trials, many are conducted in energy-intensive healthcare facilities. Traditional trial sites, such as hospitals and dedicated research centers, can have high energy demands for equipment, lighting, and climate control. These facilities often rely on regional or national power grids for their energy needs. Integrating renewable energy solutions in these facilities can also be costly and challenging, because it can involve significant investments into new equipment, renewable energy projects, and more.

In decentralized clinical trials, the reduction in infrastructure and onsite resources will allow for a lower energy demand overall. This, in turn, will result in benefits such as simplified trial designs, reduced bureaucracy, and less human travel required for video conferencing. Furthermore, the additional appointments required for clinical trials might create additional time and financial burdens for participants. Decentralized clinical trials can reduce the burden on patients for in-person visits and increase patient retention and long-term follow-up.

Core pillars on how AWS can power sustainable decentralized clinical trials

AWS customers have developed proven solutions that power sustainable decentralized clinical trials. SourceFuse is an AWS partner that has developed a mobile app and web interface that enables patients to participate in decentralized clinical trials remotely from their homes, eliminating the environmental impact of travel and paper-based data collection. The platform’s cloud-centered architecture, built on AWS services, supports the scalable and sustainable operation of these remote clinical trials.

In this post, we provide sustainability-oriented guidance focused on four key areas: virtual trials, personalized patient engagement, patient-centric trial design, and centralized data management. The following figure showcases the AWS services that can help in these four areas.

Personalized remote patient engagement

The average dropout rate for clinical trials is 30%, so providing an omnichannel experience for subjects to interact with trial facilitators is imperative. Because decentralized clinical trials provide flexibility for patients to participate at home, the experience for patients to collect and report data should be seamless. One solution is to use voice applications to enable patient data reporting, using Amazon Alexa and Amazon Connect. For example, a patient can report symptoms to their Amazon Echo device, invoking an automated patient outreach scheduler using Amazon Connect.

Trial facilitators can also use Amazon Pinpoint to connect with customers through multiple channels. They can use Amazon Pinpoint to send medication reminders, automate surveys, or push other communications without the need for paper mail delivery.

Virtual trials

Decentralized clinical trials reduce emissions compared to regular clinical trials by eliminating the need for travel and physical infrastructure. Instead, a core component of decentralized clinical trials is a secure, scalable data infrastructure with strong data analytics capabilities. Amazon Redshift is a fully managed cloud data warehouse that trial scientists can use to perform analytics.

Clinical Research Organizations (CROs) and life sciences organizations can also use AWS for mobile device and wearable data capture. Patients, in the comfort of their own home, can collect data passively through wearables, activity trackers, and other smart devices. This data is streamed to AWS IoT Core, which can write data to Amazon Data Firehose in real time. This data can then be sent to services like Amazon Simple Storage Service (Amazon S3) and AWS Glue for data processing and insight extraction.

Patient-centric trial design

A key characteristic of decentralized clinical trials is patient-centric protocol design, which prioritizes the patients’ needs throughout the entire clinical trial process. This involves patient-reported outcomes and often implement flexible participation, which can complicate protocol development and necessitate more extensive regulatory documentation. This can add days or even weeks to the lifespan of a trial, leading to avoidable costs. Amazon SageMaker enables trial developers to build and train machine learning (ML) models that reduce the likelihood of protocol amendments and inconsistencies. Models can also be built to determine the appropriate sample size and recruitment timelines.

With SageMaker, you can optimize your ML environment for sustainability. Amazon SageMaker Debugger provides profiler capabilities to detect under-utilization of system resources, which helps right-size your environment and avoid unnecessary carbon emissions. Organizations can further reduce emissions by choosing deployment regions near renewable energy projects. Currently, there are 22 AWS data center regions where 100% of the electricity consumed is matched by renewable energy sources. Additionally, you can use Amazon Q, a generative AI-powered assistant, to surface and generate potential amendments to avoid expensive costs associated with protocol revisions.

Centralized data management

CROs and bio-pharmaceutical companies are striving to achieve end-to-end data linearity for all clinical trials within an organization. They want to see traceability across the board, while achieving data harmonization for regulatory clinical trial guardrails. The pipeline approach to data management in clinical trials has led to siloed, disconnected data across an organization, because separate storage is used for each trial. Decentralized clinical trials, however, often employ a singular data lake for all of an organization’s clinical trials.

With a centralized data lake, organizations can avoid the duplication of data across separate trial databases. This leads to savings in storage costs and computing resources, as well as a reduction in the environmental impact of maintaining multiple data silos. To build a data management platform, the process could begin with ingesting and normalizing clinical trial data using AWS HealthLake. HealthLake is designed to ingest data from various sources, such as electronic health records, medical imaging, and laboratory results, and automatically transform the data into the industry-standard FHIR format. This clinical voice application solution built entirely on AWS showcases the advantages of having a centralized location for clinical data, such as avoiding data drift and redundant storage.

With the normalized data now available in HealthLake, the next step would be to orchestrate the various data processing and analysis workflows using AWS Step Functions. You can use Step Functions to coordinate the integration of the HealthLake data into a centralized data lake, as well as invoke subsequent processing and analysis tasks. This could involve using serverless computing with AWS Lambda to perform event-driven data transformation, quality checks, and enrichment activities. By combining the power powerful data normalization capabilities of HealthLake and the orchestration features of Step Functions, the platform can provide a robust, scalable, and streamlined approach to managing decentralized clinical trial data within the organization.

Conclusion

In this post, we discussed the critical importance of sustainability in clinical trials. We provided an overview of the key distinctions between traditional centralized clinical trials and decentralized clinical trials. Importantly, we explored how AWS technologies can enable the development of more sustainable clinical trials, addressing the four main pillars that underpin a successful decentralized trial approach.

To learn more about how AWS can power sustainable clinical trials for your organization, reach out to your AWS Account representatives. For more information about optimizing your workloads for sustainability, see Optimizing Deep Learning Workloads for Sustainability on AWS.

References

About the Authors

Sid Rampally is a Customer Solutions Manager at AWS driving GenAI acceleration for Life Sciences customers. He writes about topics relevant to his customers, focusing on data engineering and machine learning. In his spare time, Sid enjoys walking his dog in Central Park and playing hockey.

Nina Chen is a Customer Solutions Manager at AWS specializing in leading software companies to leverage the power of the AWS cloud to accelerate their product innovation and growth. With over 4 years of experience working in the strategic Independent Software Vendor (ISV) vertical, Nina enjoys guiding ISV partners through their cloud transformation journeys, helping them optimize their cloud infrastructure, driving product innovation, and deliver exceptional customer experiences.

Read More

Amazon Q Business is a generative AI-powered assistant designed to enhance enterprise operations. It’s a fully managed service that helps provide accurate answers to users’ questions while adhering to the security and access restrictions of the content. You can tailor Amazon Q Business to your specific business needs by connecting to your company’s information and enterprise systems using built-in connectors to a variety of enterprise data sources. It enables users in various roles, such as marketing managers, project managers, and sales representatives, to have tailored conversations, solve business problems, generate content, take action, and more, through a web interface. This service aims to help make employees work smarter, move faster, and drive significant impact by providing immediate and relevant information to help them with their tasks.

One such enterprise data repository you can use to store and manage content is Google Drive. Google Drive is a cloud-based storage service that provides a centralized location for storing digital assets, including documents, knowledge articles, and spreadsheets. This service helps your teams collaborate effectively by enabling the sharing and organization of important files across the enterprise. To use Google Drive within Amazon Q Business, you can configure the Amazon Q Business Google Drive connector. This connector allows Amazon Q Business to securely index files stored in Google Drive using access control lists (ACLs). These ACLs make sure that users only access the documents they’re permitted to view, allowing them to ask questions and retrieve information relevant to their work directly through Amazon Q Business.

This post covers the steps to configure the Amazon Q Business Google Drive connector, including authentication setup and verifying the secure indexing of your Google Drive content.

Index Google Drive documents using the Amazon Q Google Drive connector

The Amazon Q Google Drive connector can index Google Drive documents hosted in a Google Workspace account. The connector can’t index documents stored on Google Drive in a personal Google Gmail account. Amazon Q Business can authenticate with your Google Workspace using a service account or OAuth 2.0 authentication. A service account enables indexing files for user accounts across an enterprise in a Google Workspace. Using OAuth 2.0 authentication allows for crawling and indexing files in a single Google Workspace account. This post shows you how to configure Amazon Q Business to authenticate using a Google service account.

Google prescribes that in order to index multiple users’ documents, the crawler must support the capability to authenticate with a service account with domain-wide delegation. This allows the connector to index the documents of all users in your drive and shared drives. Amazon Q Business connectors only crawl the documents that the Amazon Q Business application administrator specifies need to be crawled. Administrators can specify the paths to crawl, specific file name patterns, or types. Amazon Q Business doesn’t use customer data to train any models. All customer data is indexed only in the customer account. Also, Amazon Q Business Connectors will only index content specified by the administrator. It won’t index any content on its own without explicitly being configured to do so by the administrator of Amazon Q Business.

You can configure the Amazon Q Google Drive connector to crawl and index file types supported by Amazon Q Business. Google Write documents are exported as Microsoft Word and Google Sheet documents are exported as Microsoft Excel during the crawling phase.

Metadata

Every document has structural attributes—or metadata—attached to it. Document attributes can include information such as document title, document author, time created, time updated, and document type.

When you connect Amazon Q Business to a data source, it automatically maps specific data source document attributes to fields within an Amazon Q Business index. If a document attribute in your data source doesn’t have an attribute mapping already available, or if you want to map additional document attributes to index fields, you can use the custom field mappings to specify how a data source attribute maps to an Amazon Q Business index field. You can create field mappings by editing your data source after your application and retriever are created.

There are four default metadata attributes indexed for each Google Drive document: authors, source URL, creation date, and last update date. You can also select additional reserved data field mappings.

Amazon Q Business crawls Google Drive ACLs defined in a Google Workspace for document security. Google Workspace users and groups are mapped to the _user_id and _group_ids fields associated with the Amazon Q Business application in AWS IAM Identity Center. These user and group associations are persisted in the user store associated with the Amazon Q Business index created for crawled Google Drive documents.

Overview of ACLs in Amazon Q Business

In the context of knowledge management and generative AI chatbot applications, an ACL plays a crucial role in managing who can access information and what actions they can perform within the system. They also facilitate knowledge sharing within specific groups or teams while restricting access to others.

In this solution, we deploy an Amazon Q web experience to demonstrate that two business users can only ask questions about documents they have access to according to the ACL. With the Amazon Q Business Google Drive connector, the Google Workspace ACL will be ingested with documents. This enables Amazon Q Business to control the scope of documents that each user can access in the Amazon Q web experience.

Authentication types

An Amazon Q Business application requires you to use IAM Identity Center to manage user access. Although it’s recommended to have an IAM Identity Center instance configured (with users federated and groups added) before you start, you can also choose to create and configure an IAM Identity Center instance for your Amazon Q Business application using the Amazon Q console.

You can also add users to your IAM Identity Center instance from the Amazon Q Business console, if you aren’t federating identity. When you add a new user, make sure that the user is enabled in your IAM Identity Center instance and that they have verified their email ID. They need to complete these steps before they can log in to your Amazon Q Business web experience.

Your identity source in IAM Identity Center defines where your users and groups are managed. After you configure your identity source, you can look up users or groups to grant them single sign-on access to AWS accounts, applications, or both.

You can have only one identity source per organization in AWS Organizations. You can choose one of the following as your identity source:

• IAM Identity Center directory – When you enable IAM Identity Center for the first time, it’s automatically configured with an IAM Identity Center directory as your default identity source. This is where you create your users and groups, and assign their level of access to your AWS accounts and applications. For more details, see Manage identities in IAM Identity Center.
• Active Directory – Choose this option if you want to continue managing users in either your AWS Managed Microsoft AD directory using AWS Directory Service or your self- managed directory in Active Directory (AD).
• External identity provider – Choose this option if you want to manage users in other external identity providers (IdPs) through the SAML 2.0 standard, such as Okta.
• IAM identity provider – Amazon Q Business applications can now federate with an enterprise’s IAM IdP. For more information, refer to Build private and secure enterprise generative AI applications with Amazon Q Business using IAM Federation.

Overview of solution

With Amazon Q Business, you can configure multiple data sources to provide a central place to search across your document repository. For our solution, we demonstrate how to index Google Drive data using the Amazon Q Business Google Drive connector. We complete the following steps:

1. Configure Google Workspace prerequisites.
2. Configure an Amazon Q Business application.
3. Connect Google Drive to Amazon Q Business.
4. Create users and index the data in the Google Drive.
5. Run a sample query to test the solution.

Configure Google Workspace prerequisites

For this solution, Amazon Q will connect to a Google Workspace and crawl Google Drive documents owned by business users in different groups using a service account. Complete the following steps to configure your Google Workspace:

1. Log in to the Google API console as an admin user.
2. Choose the dropdown menu next to the search box, then choose New Project.
   
   
3. Enter the project name, choose the Google organization, and choose Create.
   

The Google Drive and Admin SDK APIs need to be enabled for Amazon Q to crawl Google Drive files.

4. Search for each API on the Google Cloud console and choose Enable.
   
5. Search for Service Accounts to access the IAM & Admin navigation pane and choose Create Service Account.
6. Enter the service account name, service account ID, and description, and choose Done.
7. Choose the email of the service account created in the previous step.
8. On the Keys tab, choose Add Key, then choose Create New Key.
9. For Key type, select JSON, and choose Create to download and locally save a new private key.

Now we enable domain-wide delegation for the five required API scopes on the Domain-wide Delegation page.

10. Choose Add new.
11. Add the following comma delimited API scopes for client ID generated for the private key created in the previous step:
    https://www.googleapis.com/auth/drive.readonly,
https://www.googleapis.com/auth/drive.metadata.readonly,
https://www.googleapis.com/auth/admin.directory.group.readonly,
https://www.googleapis.com/auth/admin.directory.user.readonly,
https://www.googleapis.com/auth/cloud-platform
12. Choose Authorize.
    

Now we create users and add them to groups.

13. Navigate to the Google Workspace Admin console and choose Users in the navigation pane.
14. Choose Add new user to create two new business users.
    
15. Choose Groups in the navigation pane.
16. Choose Create group to create two Google groups and add one business user to each group.
    
17. Upload files that Amazon Q supports into each business user’s Google Drive.

In this solution, we upload the Amazon 2020 annual report to the first business user’s Google Drive and upload the Amazon 2021 annual report and Amazon 2022 annual report to the second business user’s Google Drive.

The business user that uploaded the Amazon 2021 annual report can also share it with the other business user’s Google group.

18. Choose the options menu (three vertical dots) for the Google Drive file and choose Share.
19. Enter the name of the other Google group and choose Send.

Create an Amazon Q Business application with a Google Drive connector

An Amazon Q Business application needs to be created with a Google Drive connector to crawl and index Google Drive files. To create an Amazon Q application, complete the following steps:

1. On the Amazon Q console, choose Applications in the navigation pane.
2. Choose Create application.
3. For Application name, enter a name.
4. Leave application configuration settings as defaults.
5. Choose Create.
   
6. After the application is created, choose Data Sources.
7. Then choose Select retriever and Confirm to use a Native retriever and Enterprise provisioning.
   
8. After confirming retriever settings, choose Add data source, and then choose the plus sign next to Google Drive.
   
9. Under Name and description, enter a data source name and optional description.
10. Under Authentication, select Google service account and choose Create a new secret from the AWS Secrets Manager secret drop down to create an AWS Secrets Manager secret.
    
11. Enter a secret name, admin account email, client email, and the JSON key you downloaded earlier, then choose Save.
    
12. Under IAM role, choose Create a new service role.
13. Under Additional Configuration, choose User email, and add the two recently created Google Workspace business user email addresses.
    
14. Under Sync run schedule, for Frequency, choose Run on demand.
15. Choose Add data source.
    

Create and manage users

To create an Amazon Q web experience accessible by Google Workspace users, you need to create corresponding users in IAM Identity Center. Amazon Q applications are only accessible by IAM Identity Center users with user identities that own indexed documents. To create the IAM Identity Center users, complete the following steps:

1. On the IAM Identity Center console, choose Users in the navigation pane.
2. Choose Add user.
3. Create IAM Identity Center users that mirror your Google Workspace users by entering the required user information.
   
4. Accept the IAM Identity Center invitation sent through email to each new business user and set each business user’s IAM Identity Center password.
5. On the Amazon Q Business console, navigate to the application with the Google Drive data source.
6. Choose Manage user access.
7. Choose Add groups and users, select Assign existing users and groups, and choose Next.
   
8. Assign users to the Amazon Q application, choose Assign, and choose Confirm if each business user is subscribed to Q Business Pro.
   

After you add IAM Identity Center users to your Amazon Q application, its web experience URL will appear in the Q Business applications list. You can use the URL to connect to the Amazon Q web experience with either of your Google business users. By default, each user can only ask questions about documents in their Google Drive.

Run sample queries in Amazon Q

To test the Amazon Q application with the Amazon annual reports you uploaded to Google Drive, complete the following steps:

1. On the Amazon Q Business console, navigate to the data source you created.
2. Run an on-demand sync of the data source by choosing Sync now.
   
3. Navigate to the web experience URL in a new private browser window and log in as the first business user.
   
4. Ask Amazon Q a question, such as how many employees work at Amazon.

The source documents should be the Amazon 2020 and 2021 annual reports, assuming the first business user uploaded the Amazon 2020 annual report and the second business user shared the Amazon 2021 annual report with the first business user.

5. Navigate to the web experience URL in a new private browser window and log in as the second business user.
6. Ask Amazon Q the same question (how many employees work at Amazon).

The source documents should be the Amazon 2021 and 2022 annual reports.

Troubleshooting

In this section, we share some common issues and troubleshooting tips.

IAM Identity Center login error

You might receive an error on the IAM Identity Center login page that says “We couldn’t verify your sign-in credentials.”

To troubleshoot, complete the following steps:

1. Confirm that the business users that mirror the Google Workspace users were created in IAM Identity Center.
2. If the users exist, navigate to the user in IAM Identity Center and choose Reset password, then select Generate a one-time password and share the password with the user.

A password will be provided for login and the user will be asked to change their password after a successful login.

Google Drive data source crawling or indexing failure

If the Google Drive data source crawling or indexing fails, complete the following steps:

1. Confirm the business users provisioned in the Google Workspace are members of the Google groups.
2. Inspect the Amazon CloudWatch logs for the last time the Google Drive data source was crawled for users with Google Drive files in the Google Workspace.
3. If the crawler didn’t successfully log the indexing of an expected user’s files, check the IAM Identity Center users, then compare the attributes in the Secrets Manager secret to the corresponding Google Workspace attributes, including client ID, service account email, and service account private key.
4. Use the Amazon Q Business document-level sync reports to confirm the intended Google Drive documents were indexed by Amazon Q.

Google Drive data source crawling and indexing job doesn’t crawl and index documents

If the Google Drive data source crawling and indexing job doesn’t crawl and index any documents, complete the following steps:

1. Confirm the business users provisioned in the Google Workspace are members of the Google groups.
2. Confirm there are IAM Identity Center users that mirror the Google Workspace users.
3. Confirm both IAM Identity Center users subscribe to Q Business Pro.
4. Confirm the Google Workspace admin user has enabled the Google Drive API.

Amazon Q web experience doesn’t return expected answers from the expected source

If the Amazon Q web experience doesn’t return expected answers from the expected source, complete the following steps:

1. Upload the expected source document into an Amazon Q Business chat session by choosing the paperclip icon in the Amazon Q chat interface and then choosing the file.
   

After you upload the document into the session, if the expected answers are generated from the expected document, the document wasn’t successfully indexed from the Google Drive data source.

2. If Amazon Q doesn’t return the expected answer for the uploaded document, modify the prompt used to ask the question.

Clean up

To prevent incurring additional costs, it’s essential to clean up and remove any resources created during the implementation of this solution. Specifically, you should delete the Amazon Q application, which will consequently remove the associated index and data connectors. However, any Secrets Manager secrets created during the Amazon Q application setup process need to be removed separately. Failing to clean up these resources may result in ongoing charges, so it’s crucial to take the necessary steps to completely remove all components related to this solution.

Complete the following steps to delete the Amazon Q application, secret, and IAM Identity Center users in your AWS account:

1. On the Amazon Q Business console, choose Applications in the navigation pane.
2. Select the application that you created and on the Actions menu, choose Delete and confirm the deletion.
3. On the Secrets Manager console, choose Secrets in the navigation pane.
4. Select the secret that was created for the Google Drive connector and on the Actions menu, choose Delete.
5. Specify the waiting period as 7 days and choose Schedule deletion.
6. On the IAM Identity Center console, choose Users in the navigation pane.
7. Select the two users that you created and choose Delete users to remove these users.

Additionally, you should remove the business users added to your Google Workspace during the implementation of this solution because Google Workspaces costs are billed on a per-user basis.

Conclusion

In this post, you created an Amazon Q application that indexed Google Drive documents using the Google Drive connector. You were able to connect to the Amazon Q conversational interface as each of your business users and ask questions about the documents each user could access in accordance with the ACL.

You can continue to experiment by adding more PDF documents to your business users’ Google Drives and re-syncing your Amazon Q Google Drive data source.

Amazon Q Business offers other connectors, such as for Confluence Cloud. To learn more about the Amazon Q Business Confluence Cloud connector, refer to Connecting Confluence (Cloud) to Amazon Q Business.

About the Authors

Glen Ireland is a Senior Enterprise Account Engineer at AWS in the Worldwide Public Sector. Glen’s areas of focus include empowering customers interested in building generative AI solutions using Amazon Q.

Julia Hu is a Specialist Solutions Architect who helps AWS customers and partners build generative AI solutions using Amazon Q Business on AWS. Julia has over 4 years of experience developing solutions for customers adopting AWS services on the forefront of cloud technology.

Read More

This post is co-written with David Gildea and Tom Nijs from Druva.

Druva enables cyber, data, and operational resilience for thousands of enterprises, and is trusted by 60 of the Fortune 500. Customers use Druva Data Resiliency Cloud to simplify data protection, streamline data governance, and gain data visibility and insights. Independent software vendors (ISVs) like Druva are integrating AI assistants into their user applications to make software more accessible.

Dru, the Druva backup AI copilot, enables real-time interaction and personalized responses, with users engaging in a natural conversation with the software. From finding inconsistencies and errors across the environment to scheduling backup jobs and setting retention policies, users need only ask and Dru responds. Dru can also recommend actions to improve the environment, remedy backup failures, and identify opportunities to enhance security.

In this post, we show how Druva approached natural language querying (NLQ)—asking questions in English and getting tabular data as answers—using Amazon Bedrock, the challenges they faced, sample prompts, and key learnings.

Use case overview

The following screenshot illustrates the Dru conversation interface.

In a single conversation interface, Dru provides the following:

• Interactive reporting with real-time insights – Users can request data or customized reports without extensive searching or navigating through multiple screens. Dru also suggests follow-up questions to enhance user experience.
• Intelligent responses and a direct conduit to Druva’s documentation – Users can gain in-depth knowledge about product features and functionalities without manual searches or watching training videos. Dru also suggests resources for further learning.
• Assisted troubleshooting – Users can request summaries of top failure reasons and receive suggested corrective measures. Dru on the backend decodes log data, deciphers error codes, and invokes API calls to troubleshoot.
• Simplified admin operations, with increased seamlessness and accessibility – Users can perform tasks like creating a new backup policy or triggering a backup, managed by Druva’s existing role-based access control (RBAC) mechanism.
• Customized website navigation through conversational commands – Users can instruct Dru to navigate to specific website locations, eliminating the need for manual menu exploration. Dru also suggests follow-up actions to speed up task completion.

Challenges and key learnings

In this section, we discuss the challenges and key learnings of Druva’s journey.

Overall orchestration

Originally, we adopted an AI agent approach and relied on the foundation model (FM) to make plans and invoke tools using the reasoning and acting (ReAct) method to answer user questions. However, we found the objective too broad and complicated for the AI agent. The AI agent would take more than 60 seconds to plan and respond to a user question. Sometimes it would even get stuck in a thought-loop, and the overall success rate wasn’t satisfactory.

We decided to move to the prompt chaining approach using a directed acyclic graph (DAG). This approach allowed us to break the problem down into multiple steps:

1. Identify the API route.
2. Generate and invoke private API calls.
3. Generate and run data transformation Python code.

Each step became an independent stream, so our engineers could iteratively develop and evaluate the performance and speed until they worked well in isolation. The workflow also became more controllable by defining proper error paths.

Stream 1: Identify the API route

Out of the hundreds of APIs that power Druva products, we needed to match the exact API the application needs to call to answer the user question. For example, “Show me my backup failures for the past 72 hours, grouped by server.” Having similar names and synonyms in API routes make this retrieval problem more complex.

Originally, we formulated this task as a retrieval problem. We tried different methods, including k-nearest neighbor (k-NN) search of vector embeddings, BM25 with synonyms, and a hybrid of both across fields including API routes, descriptions, and hypothetical questions. We found that the simplest and most accurate way was to formulate it as a classification task to the FM. We curated a small list of examples in question-API route pairs, which helped improve the accuracy and make the output format more consistent.

Stream 2: Generate and invoke private API calls

Next, we API call with the correct parameters and invoke it. FM hallucination of parameters, particularly those with free-form JSON object, is one of the major challenges in the whole workflow. For example, the unsupported key server can appear in the generated parameter:

"filter": {
    "and": [
        {
            "gte": {
                "key": "dt",
                "value": 1704067200
            }
        },
        {
            "eq": {
                "key": "server",
                "value": "xyz"
            }
        }
    ]
}

We tried different prompting techniques, such as few-shot prompting and chain of thought (CoT), but the success rate was still unsatisfactory. To make API call generation and invocation more robust, we separated this task into two steps:

1. First, we used an FM to generate parameters in a JSON dictionary instead of a full API request headers and body.
2. Afterwards, we wrote a postprocessing function to remove parameters that didn’t conform to the API schema.

This method provided a successful API invocation, at the expense of getting more data than required for downstream processing.

Stream 3: Generate and run data transformation Python code

Next, we took the response from the API call and transformed it to answer the user question. For example, “Create a pandas dataframe and group it by server column.” Similar to stream 2, FM hallucination is again an obstacle. Generated code can contain syntax errors, such as confusing PySpark functions with Pandas functions.

After trying many different prompting techniques without success, we looked at the reflection pattern, asking the FM to self-correct code in a loop. This improved the success rate at the expense of more FM invocations, which were slower and more expensive. We found that although smaller models are faster and more cost-effective, at times they had inconsistent results. Anthropic’s Claude 2.1 on Amazon Bedrock gave more accurate results on the second try.

Model choices

Druva selected Amazon Bedrock for several compelling reasons, with security and latency being the most important. A key factor in this decision was the seamless integration with Druva’s services. Using Amazon Bedrock aligned naturally with Druva’s existing environment on AWS, maintaining a secure and efficient extension of their capabilities.

Additionally, one of our primary challenges in developing Dru involved selecting the optimal FMs for specific tasks. Amazon Bedrock effectively addresses this challenge with its extensive array of available FMs, each offering unique capabilities. This variety enabled Druva to conduct the rapid and comprehensive testing of various FMs and their parameters, facilitating the selection of the most suitable one. The process was streamlined because Druva didn’t need to delve into the complexities of running or managing these diverse FMs, thanks to the robust infrastructure provided by Amazon Bedrock.

Through the experiments, we found that different models performed better in specific tasks. For example, Meta Llama 2 performed better with code generation task; Anthropic Claude Instance was good in efficient and cost-effective conversation; whereas Anthropic Claude 2.1 was good in getting desired responses in retry flows.

These were the latest models from Anthropic and Meta at the time of this writing.

Solution overview

The following diagram shows how the three streams work together as a single workflow to answer user questions with tabular data.

The following are the steps of the workflow:

1. The authenticated user submits a question to Dru, for example, “Show me my backup job failures for the last 72 hours,” as an API call.
2. The request arrives at the microservice on our existing Amazon Elastic Container Service (Amazon ECS) cluster. This process consists of the following steps:
   1. A classification task using the FM provides the available API routes in the prompt and asks for the one that best matches with user question.
   2. An API parameters generation task using the FM gets the corresponding API swagger, then asks the FM to suggest key-value pairs to the API call that can retrieve data to answer the question.
   3. A custom Python function verifies, formats, and invokes the API call, then passes the data in JSON format to the next step.
   4. A Python code generation task using the FM samples a few records of data from the previous step, then asks the FM to write Python code to transform the data to answer the question.
   5. A custom Python function runs the Python code and returns the answer in tabular format.

To maintain user and system security, we make sure in our design that:

• The FM can’t directly connect to any Druva backend services.
• The FM resides in a separate AWS account and virtual private cloud (VPC) from the backend services.
• The FM can’t initiate actions independently.
• The FM can only respond to questions sent from Druva’s API.
• Normal customer permissions apply to the API calls made by Dru.
• The call to the API (Step 1) is only possible for authenticated user. The authentication component lives outside the Dru solution and is used across other internal solutions.
• To avoid prompt injection, jailbreaking, and other malicious activities, a separate module checks for these before the request reaches this service (Amazon API Gateway in Step 1).

For more details, refer to Druva’s Secret Sauce: Meet the Technology Behind Dru’s GenAI Magic.

Implementation details

In this section, we discuss Steps 2a–2e in the solution workflow.

2a. Look up the API definition

This step uses an FM to perform classification. It takes the user question and a full list of available API routes with meaningful names and descriptions as the input, and responds The following is a sample prompt:

Please read the following API routes carefully as I’ll ask you a question about them:
<api_routes>{api_routes}</api_routes>
Which API route can best answer “{question}”?

2b. Generate the API call

This step uses an FM to generate API parameters. It first looks up the corresponding swagger for the API route (from Step 2a). Next, it passes the swagger and the user question to an FM and responds with some key-value pairs to the API route that can retrieve relevant data. The following is a sample prompt:

Please read the following swagger carefully as I’ll ask you a question about it:
<swagger>{swagger}</swagger>
Produce a key-value JSON dict of the available request parameters based on “{question}” with reference to the swagger.

2c. Validate and invoke the API call

In the previous step, even with an attempt to ground responses with swagger, the FM can still hallucinate wrong or nonexistent API parameters. This step uses a programmatic way to verify, format, and invoke the API call to get data. The following is the pseudo code:

for each input parameter (key/value)
  if parameter key not in swagger then
    drop parameter
  else if parameter value data type not match swagger then
    drop parameter
  else
    URL encode parameter
  end if
end for

2d. Generate Python code to transform data

This step uses an FM to generate Python code. It first samples a few records of input data to reduce input tokens. Then it passes the sample data and the user question to an FM and responds with a Python script that transforms data to answer the question. The following is a sample prompt:

Please read the following sample data carefully as I’ll ask you a question about them:
<sample_data>{5_rows_of_data_in_json}</sample_data>
Write a Python script using pandas to transform the data to answer the question “{question}”.

2e. Run the Python code

This step involves a Python script, which imports the generated Python package, runs the transformation, and returns the tabular data as the final response. If an error occurs, it will invoke the FM to try to correct the code. When everything fails, it returns the input data. The following is the pseudo code:

for maximum number of retries
  run data transformation function
  if error then
    invoke foundation model to correct code
  end if
end for
if success then
  return transformed data
else
  return input data
end if

Conclusion

Using Amazon Bedrock for the solution foundation led to remarkable achievements in accuracy, as evidenced by the following metrics in our evaluations using an internal dataset:

• Stream 1: Identify the API route – Achieved a perfect accuracy rate of 100%
• Stream 2: Generate and invoke private API calls – Maintained this standard with a 100% accuracy rate
• Stream 3: Generate and run data transformation Python code – Attained a highly commendable accuracy of 90%

These results are not just numbers; they are a testament to the robustness and efficiency of the Amazon Bedrock based solution. With such high levels of accuracy, Druva is now poised to confidently broaden their horizons. Our next goal is to extend this solution to encompass a wider range of APIs across Druva products. The next expansion will be scaling up usage and substantially enrich the experience of Druva customers. By integrating more APIs, Druva will offer a more seamless, responsive, and contextual interaction with Druva products, further enhancing the value delivered to Druva users.

To learn more about Druva’s AI solutions, visit the Dru solution page, where you can see some of these capabilities in action through recorded demos. Visit the AWS Machine Learning blog to see how other customers are using Amazon Bedrock to solve their business problems.

About the Authors

David Gildea is the VP of Product for Generative AI at Druva. With over 20 years of experience in cloud automation and emerging technologies, David has led transformative projects in data management and cloud infrastructure. As the founder and former CEO of CloudRanger, he pioneered innovative solutions to optimize cloud operations, later leading to its acquisition by Druva. Currently, David leads the Labs team in the Office of the CTO, spearheading R&D into generative AI initiatives across the organization, including projects like Dru Copilot, Dru Investigate, and Amazon Q. His expertise spans technical research, commercial planning, and product development, making him a prominent figure in the field of cloud technology and generative AI.

Tom Nijs is an experienced backend and AI engineer at Druva, passionate about both learning and sharing knowledge. With a focus on optimizing systems and using AI, he’s dedicated to helping teams and developers bring innovative solutions to life.

Corvus Lee is a Senior GenAI Labs Solutions Architect at AWS. He is passionate about designing and developing prototypes that use generative AI to solve customer problems. He also keeps up with the latest developments in generative AI and retrieval techniques by applying them to real-world scenarios.

Fahad Ahmed is a Senior Solutions Architect at AWS and assists financial services customers. He has over 17 years of experience building and designing software applications. He recently found a new passion of making AI services accessible to the masses.

Read More