Monthly Archives: July 2024

In today’s digital landscape, the demand for audio and video content is skyrocketing. Organizations are increasingly using media to engage with their audiences in innovative ways. From product documentation in video format to podcasts replacing traditional blog posts, content creators are exploring diverse channels to reach a wider audience. The rise of virtual workplaces has also led to a surge in content captured through recorded meetings, calls, and voicemails. Additionally, contact centers generate a wealth of media content, including support calls, screen-share recordings, and post-call surveys.

We are excited to introduce Mediasearch Q Business, an open source solution powered by Amazon Q Business and Amazon Transcribe. Mediasearch Q Business builds on the Mediasearch solution powered by Amazon Kendra and enhances the search experience using Amazon Q Business. Mediasearch Q Business supercharges the way you consume media files by using them as part of the knowledge base used by Amazon Q Business to generate reliable answers to user questions. The solution also features an enhanced Amazon Q Business query application that allows users to play the relevant section of the original media files or YouTube videos directly from the search results page, providing a seamless and intuitive user experience.

Solution overview

Mediasearch Q Business is straightforward to install and try out.

The solution has two components, as illustrated in the following diagram:

• A Mediasearch indexer that transcribes media files (audio and video) on an Amazon Simple Storage Service (Amazon S3) bucket or media from a YouTube playlist and ingests the transcriptions into either an Amazon Q Business native index (configured as part of the Amazon Q Business application) or an Amazon Kendra
• A Mediasearch finder, which provides a UI and makes API calls to the Amazon Q Business service APIs on behalf of the logged-in user. The response from API calls are displayed to the end-user.

The Mediasearch indexer finds and transcribes audio and video files stored in an S3 bucket. The indexer can also index YouTube videos from a YouTube playlist as audio files and transcribe these audio files. It prepares the transcriptions by embedding time markers at the start of each sentence, and it indexes each prepared transcription in an Amazon Q Business native retriever or an Amazon Kendra retriever. The indexer runs the first time when you install it, and subsequently runs on an interval that you specify, maintaining the index to reflect any new, modified, or deleted files.

The Mediasearch finder is a web search client that you use to search for content in your Amazon Q Business application. Additionally, the Mediasearch finder includes in-line embedded media players in the search result, so you can see the relevant section of the transcript, and play the corresponding section from the original media (audio files and video files in your media bucket or a YouTube video) without navigating away from the search page.

In the sections that follow, we discuss the following topics:

• How to deploy the solution to your AWS account
• How to use it to index and search sample media files
• How to use the solution with your own media files
• How the solution works
• The estimated costs involved
• How to monitor usage and troubleshoot problems
• Options to customize and tune the solution
• How to uninstall and clean up when you’re done experimenting

Prerequisites

Make sure you have the following:

• An AWS account where you can launch an AWS CloudFormation stack
• An AWS IAM Identity Center instance ARN that would be used by the Amazon Q Business application to provide access to users

Deploy the Mediasearch Q Business solution

In this section, we walk through deploying the two solution components: the indexer and the finder. We use a CloudFormation stack to deploy the necessary resources in the us-east-1 AWS Region.

If you’re deploying the solution to another Region, follow the instructions in the README available in the Mediasearch Q Business GitHub repository.

Deploy the Mediasearch Q Business indexer component

To deploy the indexer component, complete the following steps:

1. Choose Launch Stack.
   
2. In the Identity center ARN and Retriever selection section, for IdentityCenterInstanceArn, enter the ARN for your IAM Identity Center instance.

You can find the ARN on the Settings page of the IAM Identity Center console. The ARN is a required field.

3. Use default values for all other parameters. We will customize these values later to suit your specific requirements.
4. Acknowledge that the stack might create IAM resources with custom names, then choose Create stack.

The indexer stack takes around 10 minutes to deploy. Wait for the indexer to finish deploying before you deploy the Mediasearch Q Business finder.

Deploy the Mediasearch Q Business finder component

The Mediasearch finder uses Amazon Cognito to authenticate users to the solution. For an authenticated user to interact with an Amazon Q Business application, you must configure an IAM Identity Center customer managed application that either supports SAML 2.0 or OAuth 2.0.

In this post, we create a customer managed application that supports OAuth 2.0, a secure way for applications to communicate and share user data without exposing passwords. We use a technique called trusted identity propagation, which allows the Mediasearch Q Business finder app to access the Amazon Q service securely without sharing passwords between the two identity providers (Amazon Cognito and IAM Identity Center in our example).

Instead of sharing passwords, trusted identity propagation uses tokens. Tokens are like digital certificates that prove who the user is and what they’re allowed to do. AWS managed applications that work with trusted identity propagation get tokens directly from IAM Identity Center. IAM Identity Center can also exchange identity tokens and access tokens from external authorization servers like Amazon Cognito. This lets an application authenticate users and obtain tokens outside of AWS (like with Amazon Cognito, Microsoft Entra ID, or Okta), exchange that token for an IAM Identity Center token, and then use the new token to request access to AWS services like Amazon Q Business.

For more information, see Using trusted identity propagation with customer managed applications.

When the IAM Identity Center instance is in the same account where you are deploying the Mediasearch Q Business solution, the finder stack allows you to automatically create the IAM Identity Center customer managed application as part of the stack deployment.

If you use the organization instance of IAM Identity Center enabled in your management account, then you will be deploying the Mediasearch Q Business finder stack in a different AWS account. In this case, follow the steps in the README to create an IAM Identity Center application manually.

To deploy the finder component and create the IAM Identity Center customer managed application, complete the following steps:

1. Choose Launch Stack.
   
2. For IdentityCenterInstanceArn, enter the ARN for the IAM Identity Center instance. This is the same value you used while deploying the indexer stack.
3. For CreateIdentityCenterApplication, choose Yes to create the IAM Identity Center application for the Mediasearch finder application.
4. Under Mediasearch Indexer parameters, enter the Amazon Q Business application ID that was created by the indexer stack. You can copy this from the QBusinessApplicationId output of the indexer stack.
5. Select the retriever type that was used to deploy the Mediasearch indexer. (If you deployed an Amazon Kendra index, then select Kendra, otherwise select Native.
6. If you selected Kendra, enter the Amazon Kendra index ID that was used by the indexer stack.
7. For MediaBucketNames, use the MediaBucketsUsed output from the indexer CloudFormation stack to allow the search page to access media files across YTMediaBucket and Mediabucket.
8. Acknowledge that the stack might create IAM resources with custom names, then choose Create stack.

Configure user access to Amazon Q Business

To access the Mediasearch Q Business solution, add a user with an appropriate subscription to the Amazon Q Business application and to the IAM Identity Center customer managed application.

Add a user to the Amazon Q Business application

To start using the Amazon Q Business application, you can add users or groups to the Amazon Q Business application from your IAM Identity Center instance. Complete the following steps to add a user to the application:

1. Access the Amazon Q Business application by choosing the link for QBusinessApplication in the indexer CloudFormation stack outputs.
   
2. Under Groups and users, on the Users tab, choose Manage access and subscription.
   
3. Choose Add groups and users.
   
4. Choose Add existing users and groups.
   
5. Search for an existing user, choose the user, and choose Assign.
   
6. Select the added user and on the Change subscription menu, choose Update subscription tier.
   
7. Select the appropriate subscription tier and choose Confirm.

For details of each Amazon Q subscription, refer to Amazon Q Business pricing.

Assign users to the IAM Identity Center customer managed application

Now you can assign users or groups to the IAM Identity Center customer managed application. Complete the following steps to add a user:

1. From the outputs section of the finder CloudFormation stack, choose the URL for IdentityCenterApplicationConsoleURLto navigate to the customer managed application.
   

2. Choose Assign users and groups.
   

3. Select users and choose Assign users.
   

This concludes the user access configuration to the Mediasearch Q Business solution.

Test with the sample media files

When the Mediasearch indexer and finder stack are deployed, the indexer should have completed processing the audio (mp3) files for the YouTube videos and sample media files (selected AWS Podcast episodes and AWS Knowledge Center videos). You can now run your first Mediasearch query.

1. To log in to the Mediasearch finder application, choose the URL for MediasearchFinderURL in the stack outputs.
   

The Mediasearch finder application in your browser will show a splash page for Amazon Q Business.

2. Choose Get Started to access the Amazon Cognito page.
   

To access Mediasearch Q Business, you need to log in to the application using a user ID in the Amazon Cognito user pool created by the finder stack. The email address in Amazon Cognito must match the email address for the user in IAM Identity Center. Alternatively, the Mediasearch solution allows you to create a user through the application.

3. On the Create Account tab, enter your email (which matches the email address in IAM Identity Center), followed by a password and password confirmation, and choose Create Account.

Amazon Cognito will send an email with a confirmation code for email verification.

4. Enter this confirmation code to complete your email verification.
   

5. After email verification, you will now be able to log in to the Mediasearch Q Business application.
6. After you’re logged in, in the Enter a prompt box, write a query, such as “What is AWS Fargate?”

The query returns a response from Amazon Q Business based on the media (sample media files and YouTube audio sources) ingested into the index.

The response includes citations, with reference to sources. Users can verify their answer from Amazon Q Business by playing media files from their S3 buckets or YouTube starting at the time marker where the relevant information is found.

7. Use the embedded video player to play the original video inline. Observe that the media playback starts at the relevant section of the video based on the time marker.
8. To play the video full screen in a new browser tab, use the Full screen menu option in the player, or choose the media file hyperlink shown above the answer text.
9. Choose (right-click) the video file hyperlink, copy the URL, and enter it into a text editor.

If the media is an audio file for a YouTube video, it looks something like the following:

https://www.youtube.com/watch?v=unFVfqj9cQ8&t=36.58s

If the media file is a non-YouTube audio file that resides in MediaBucket, the URL looks like the following:

https://mediasearchtest.s3.amazonaws.com/mediasamples/What_is_an_Interface_VPC_Endpoint_and_how_can_I_create_Interface_Endpoint_for_my_VPC_.mp4?AWSAccessKeyId=ASIAXMBGHMGZLSYWJHGD&Expires=1625526197&Signature=BYeOXOzT585ntoXLDoftkfS4dBU%3D&x-amz-security-token=.... #t=253.52

This is a presigned S3 URL that provides your browser with temporary read access to the media file referenced in the search result. Using presigned URLs means you don’t need to provide permanent public access to all of your indexed media files.

11. Experiment with additional queries, such as “How has AWS helped customers in building MLOps platform?” or “How can I use Generative AI to improve customer experience?” or try your own questions.

Index and search your own media files

To index media files stored in your own S3 bucket, replace the MediaBucket and MediaFolderPrefix parameters with your own bucket name and prefix when you install or update the indexer component stack, and modify the MediaBucketName parameter with your own bucket name when you install or update the finder component stack. Additionally, you can replace the YouTube playlist (PlayListURL) with your own playlist URL and update the indexer stack.

1. When creating a new MediaSearch indexer stack, you can choose to use either a native retriever or an Amazon Kendra retriever. You can make this selection using the parameter RetrieverType. When using the Amazon Kendra retriever, you can either let indexer stack create an Amazon Kendra index or use an existing Amazon Kendra IndexId to add files stored in the new location. To deploy a new indexer, follow the steps from earlier in this post, but replace the defaults to specify the media bucket name and prefix for your own media files or replace the YouTube playlist URL with your own playlist URL. Make sure that you comply with the YouTube Terms of Service.
2. Alternatively, update an existing MediaSearch indexer stack to replace the previously indexed files with files from the new location or update the YouTube playlist URL or the number of videos to download from the playlist:
   1. Select the stack on the AWS CloudFormation console, choose Update, then Use current template, then Next.
   2. Modify the media bucket name and prefix parameter values as needed.
   3. Modify the YouTube Playlist URL and Number of YouTube Videos values as needed.
   4. Choose Next twice, select the acknowledgement check box, and choose Update stack.
3. Update an existing MediaSearch finder stack to change bucket names or add additional bucket names to the MediaBucketNames

When the MediaSearch indexer stack is successfully created or updated, the indexer automatically finds, transcribes, and indexes the media files stored in your S3 bucket. When it’s complete, you can submit queries and find answers from the audio tracks of your own audio and video files.

You have the option to provide metadata for any or all of your media files. Use metadata to assign values to index attributes for sorting, filtering, and faceting your search results, or to specify access control lists to govern access to the files. Metadata files can be in the same S3 folder as your media files (default), or in a parallel folder structure specified by the optional indexer parameter MetadataFolderPrefix. For more information about how to create metadata files, see Amazon S3 document metadata.

You can also provide customized transcription options for any or all of your media files. This allows you to take full advantage of Amazon Transcribe features such as custom vocabularies, automatic content redaction, and custom language models.

How the Mediasearch solution works

Let’s take a quick look at how the solution works, as illustrated in the following diagram.

The Mediasearch solution has an event-driven serverless computing architecture with the following steps:

1. You provide an S3 bucket containing the audio and video files you want to index and search. This is also known as the MediaBucket. Leave this blank if you don’t want to index media from your MediaBucket.
2. You also provide your YouTube playlist URL and the number of videos to index from the YouTube playlist. Make sure that you comply with the YouTube Terms of Service. The YTIndexer will index the latest files from the YouTube playlist. For example, if the number of videos is set to 5, then the YTIndexer will index the five latest videos in the playlist. Any YouTube video indexed prior is ignored from being indexed.
3. An AWS Lambda function fetches the YouTube videos from the playlist as audio (mp3 files) into the YTMediaBucket and also creates a metadata file in the MetadataFolderPrefix location with metadata for the YouTube video. The YouTube videoid along with the related metadata are recorded in an Amazon DynamoDB table (YTMediaDDBQueueTable).
4. Amazon EventBridge generates events on a repeating interval (every 2 hours, every 6 hours, and so on) These events invoke the Lambda function S3CrawlLambdaFunction.
5. An AWS Lambda function is invoked initially when the CloudFormation stack is first deployed, and then subsequently by the scheduled events from EventBridge. The S3CrawlLambdaFunction function crawls through the MediaBucket and the YTMediabucket and starts an Amazon Q Business index (or Amazon Kendra) data source sync job. The Lambda function lists all the supported media files (FLAC, MP3, MP4, Ogg, WebM, AMR, or WAV) and associated metadata and transcribe options stored in the user provided S3 bucket.
6. Each new file is added to another DynamoDB tracking table and submitted to be transcribed by an Amazon Transcribe job. Any file that has been previously transcribed is submitted for transcription again only if it has been modified since it was previously transcribed, or if associated Amazon Transcribe options have been updated. The DynamoDB table is updated to reflect the transcription status and last modified timestamp of each file. Any tracked files that no longer exist in the S3 bucket are removed from the DynamoDB table and from the Amazon Q Business index (or Amazon Kendra index). If no new or updated files are discovered, the Amazon Q Business index (or Amazon Kendra) data source sync job is immediately stopped. The DynamoDB table holds a record for each media file with attributes to track transcription job names and status, and last modified timestamps.
7. As each Amazon Transcribe job completes, EventBridge generates a job complete event, which invokes another Lambda function (S3JobCompletionLambdaFunction).
8. The Lambda function processes the transcription job output, generating a modified transcription that has a time marker inserted at the start of each sentence. This modified transcription is indexed in Amazon Q Business (or Amazon Kendra), and the job status for the file is updated in the DynamoDB table. When the last file has been transcribed and indexed, the Amazon Q Business (or Amazon Kendra) data source sync job is stopped.
9. The index is populated and kept in sync with the transcriptions of all the media files in the S3 bucket monitored by the Mediasearch indexer component, integrated with any additional content from any other provisioned data sources. The media transcriptions are used by the Amazon Q Business application, which allows users to find content and answers to their questions.
10. The sample finder client application enhances users’ search experience by embedding an inline media player with each source or citation that is based on a transcribed media file. The client uses the time markers embedded in the transcript to start media playback at the relevant section of the original media file.
11. An Amazon Cognito user pool is used to authenticate users and is configured to exchange tokens from IAM Identity Center to support Amazon Q Business service calls.

Estimated costs

In addition to Amazon S3 costs associated with storing your media, the Mediasearch solution incurs usage costs from the Amazon Q, Amazon Kendra (if using an Amazon Kendra index), Amazon Transcribe, and Amazon API Gateway. Additional minor costs are incurred by the other services mentioned after free tier allowances have been used. For more information, see the pricing pages for Amazon Q Business, Amazon Kendra, Amazon Transcribe, Lambda, DynamoDB, and EventBridge.

Monitor and troubleshoot

To see the details of each media file transcript job, navigate to the Transcription jobs page on the Amazon Transcribe console.

Each media file is transcribed only one time, unless the file is modified. Modified files are re-transcribed and re-indexed to reflect the changes.

Choose any transcription job to review the transcription and examine additional job details.

You can check the status of the data source sync by navigating to the Amazon Q Business application deployed by the indexer stack (choose the link on the indexer stack outputs page for QApplication). In the data source section, choose the custom data source and view the status of the sync job.

On the DynamoDB console, choose Tables in the navigation pane. Use your MediaSearch stack name as a filter to display the MediaSearch DynamoDB tables, and examine the items showing each indexed media file and corresponding status.

The table MediaSearch-Indexer-YTMediaDDBQueueTable has one record for each YouTube videoid that is downloaded as an audio (mp3) file along with the metadata for the video like author, view count, video title, and so on.

The table MediaSearch-Indexer-MediaDynamoTable has one record for each media file (including YouTube videos), and contains attributes with information about the file and its processing status.

On the Functions page of the Lambda console, use your indexer stack name as a filter to list the Lambda functions that are part of the solution:

• The YouTubeVideoIndexer function indexes and downloads YouTube videos if the CloudFormation stack parameter PlayListURL is set to a valid YouTube playlist
• The S3CrawlLambdaFunction function crawls the YTMediaBucket and the MediaBucket for media files and initiates the transcription jobs for the media files

When the transcription job is complete, a completion event invokes the S3JobCompletionLambdaFunction function, which ingests the transcription into the Amazon Q Business index (or Amazon Kendra index) with any related metadata.

Choose any of the functions to examine the function details, including environment variables, source code, and more. Choose Monitor and View logs in CloudWatch to examine the output of each function invocation and troubleshoot any issues.

On the Functions page of the Lambda console, use your finder stack name as a filter to list the Lambda functions that are part of the solution:

• The BuildTriggerLambda function runs the build of the finder AWS Amplify application after cloning the AWS CodeCommit repository with the finder ReactJS code.
• The IDCTokenCreateLambda function uses the authorization header that contains a JWT token from a successful authentication with Amazon Cognito to exchange bearer tokens from IAM Identity Center.
• The IDCAppCreateLambda function creates an OAuth 2.0 IAM Identity Center application to exchange tokens from IAM Identity Center and a trusted token issuer for the Amazon Cognito user pool.
• The UserConversationLambda function is called from API Gateway to list or delete Amazon Q Business conversations.
• The UserPromptsLambda function is called from API Gateway to call the chat_sync API of the Amazon Q Business service.
• The PreSignedURLCreateLambda function is called from API Gateway to create a presigned URL for S3 buckets. The presigned URL is used to play the media files residing on the Mediabucket that serves as the source for an Amazon Q Business response.

Choose any of the functions to examine the function details, including environment variables, source code, and more. Choose Monitor and View logs in CloudWatch to examine the output of each function invocation and troubleshoot any issues.

Customize and enhance the solution

You can fork the MediaSearch Q Business GitHub repository, enhance the code, and send us pull requests so we can incorporate and share your improvements.

The following are a few suggestions for features you might want to implement:

• Enhance the indexer stack to allow the existing Amazon Q Business application IDs to be used
• Extend your search sources to include other video streaming platforms relevant to your organization
• Build Amazon CloudWatch metrics and dashboards to improve the manageability of MediaSearch

Clean up

When you’re finished experimenting with this solution, clean up your resources by using the AWS CloudFormation console to delete the indexer and finder stacks that you deployed. This deletes all the resources that were created by deploying the solution.

Preexisting Amazon Q Business applications or indexes or IAM Identity Center applications or trusted token issuers that were created manually aren’t deleted.

Conclusion

The combination of Amazon Q Business and Amazon Transcribe enables a scalable, cost-effective solution to surface insights from your media files. You can use the content of your media files to find accurate answers to your users’ questions, whether they’re from text documents or media files, and consume them in their native format. This solution enhances the overall experience of the previous Mediasearch solution by using the powerful generative artificial intelligence (AI) capabilities of Amazon Q Business.

The sample MediaSearch Q Business solution is provided as open source—use it as a starting point for your own solution, and help us make it better by contributing back fixes and features through GitHub pull requests. For expert assistance, AWS Professional Services and other Amazon partners are here to help.

We’d love to hear from you. Let us know what you think in the comments section, or use the issues forum in the MediaSearch Q Business GitHub repository.

About the Authors

Roshan Thomas is a Senior Solutions Architect at Amazon Web Services. He is based in Melbourne, Australia, and works closely with power and utilities customers to accelerate their journey in the cloud. He is passionate about technology and helping customers architect and build solutions on AWS.

Anup Dutta is a Solutions Architect with AWS based in Chennai, India. In his role at AWS, Anup works closely with startups to design and build cloud-centered solutions on AWS.

Bob Strahan is a Principal Solutions Architect in the AWS Language AI Services team.

Abhinav Jawadekar is a Principal Solutions Architect in the Amazon Q Business service team at AWS. Abhinav works with AWS customers and partners to help them build generative AI solutions on AWS.

Read More

This post is co-written with Benjamin Moody from Monks.

Monks is the global, purely digital, unitary operating brand of S4Capital plc. With a legacy of innovation and specialized expertise, Monks combines an extraordinary range of global marketing and technology services to accelerate business possibilities and redefine how brands and businesses interact with the world. Its integration of systems and workflows delivers unfettered content production, scaled experiences, enterprise-grade technology and data science fueled by AI—managed by the industry’s best and most diverse digital talent—to help the world’s trailblazing companies outmaneuver and outpace their competition.

Monks leads the way in crafting cutting-edge brand experiences. We shape modern brands through innovative and forward-thinking solutions. As brand experience experts, we harness the synergy of strategy, creativity, and in-house production to deliver exceptional results. Tasked with using the latest advancements in AWS services and machine learning (ML) acceleration, our team embarked on an ambitious project to revolutionize real-time image generation. Specifically, we focused on using AWS Inferentia2 chips with Amazon SageMaker to enhance the performance and cost-efficiency of our image generation processes..

Initially, our setup faced significant challenges regarding scalability and cost management. The primary issues were maintaining consistent inference performance under varying loads, while providing generative experience for the end-user. Traditional compute resources were not only costly but also failed to meet the low latency requirements. This scenario prompted us to explore more advanced solutions from AWS that could offer high-performance computing and cost-effective scalability.

The adoption of AWS Inferentia2 chips and SageMaker asynchronous inference endpoints emerged as a promising solution. These technologies promised to address our core challenges by significantly enhancing processing speed (AWS Inferentia2 chips were four times faster in our initial benchmarks) and reducing costs through fully managed auto scaling inference endpoints.

In this post, we share how we used AWS Inferentia2 chips with SageMaker asynchronous inference to optimize the performance by four times and achieve a 60% reduction in cost per image for our real-time diffusion AI image generation.

Solution overview

The combination of SageMaker asynchronous inference with AWS Inferentia2 allowed us to efficiently handle requests that had large payloads and long processing times while maintaining low latency requirements. A prerequisite was to fine-tune the Stable Diffusion XL model with domain-specific images which were stored in Amazon Simple Storage Service (Amazon S3). For this, we used Amazon SageMaker JumpStart. For more details, refer to Fine-Tune a Model.

The solution workflow consists of the following components:

• Endpoint creation – We created an asynchronous inference endpoint using our existing SageMaker models, using AWS Inferentia2 chips for higher price/performance.
• Request handling – Requests were queued by SageMaker upon invocation. Users submitted their image generation requests, where the input payload was placed in Amazon S3. SageMaker then queued the request for processing.
• Processing and output – After processing, the results were stored back in Amazon S3 in a specified output bucket. During periods of inactivity, SageMaker automatically scaled the instance count to zero, significantly reducing costs because charges only occurred when the endpoint was actively processing requests.
• Notifications – Completion notifications were set up through Amazon Simple Notification Service (Amazon SNS), notifying users of success or errors.

The following diagram illustrates our solution architecture and process workflow.

In the following sections, we discuss the key components of the solution in more detail.

SageMaker asynchronous endpoints

SageMaker asynchronous endpoints queue incoming requests to process them asynchronously, which is ideal for large inference payloads (up to 1 GB) or inference requests with long processing times (up to 60 minutes) that need to be processed as requests arrive. The ability to serve long-running requests enabled Monks to effectively serve their use case. Auto scaling the instance count to zero allows you to design cost-optimal inference in response to spiky traffic, so you only pay for when the instances are serving traffic. You can also scale the endpoint instance count to zero in the absence of outstanding requests and scale back up when new requests arrive.

To learn how to create a SageMaker asynchronous endpoint, attach auto scaling policies, and invoke an asynchronous endpoint, refer to Create an Asynchronous Inference Endpoint.

AWS Inferentia2 chips, which powered the SageMaker asynchronous endpoints, are AWS AI chips optimized to deliver high performance for deep learning inference applications at lowest cost. Integrated within SageMaker asynchronous inference endpoints, AWS Inferentia2 chips support scale-out distributed inference with ultra-high-speed connectivity between chips. This setup was ideal for deploying our large-scale generative AI model across multiple accelerators efficiently and cost-effectively.

In the context of our high-profile nationwide campaign, the use of asynchronous computing was key in managing peak and unexpected spikes in concurrent requests to our inference infrastructure, which was expected to be in the hundreds of concurrent requests per second. Asynchronous inference endpoints, like those provided by SageMaker, offer dynamic scalability and efficient task management.

The solution offered the following benefits:

• Efficient handling of longer processing times – SageMaker asynchronous inference endpoints are perfect for scenarios where each request might involve substantial computational work. These fully managed endpoints queue incoming inference requests and process them asynchronously. This method was particularly advantageous in our application, because it allowed the system to manage fluctuating demand efficiently. The ability to process requests asynchronously makes sure our infrastructure can handle large unexpected spikes in traffic without causing delays in response times.
• Cost-effective resource utilization – One of the most significant advantages of using asynchronous inference endpoints is their impact on cost management. These endpoints can automatically scale the compute resources down to zero in periods of inactivity, without the risk of dropping or losing requests as resources scale back up.

Custom scaling policies using Amazon CloudWatch metrics

SageMaker endpoint auto scaling behavior is defined through the use of a scaling policy, which helps us scale to multiple users using the application concurrently This policy defines how and when to scale resources up or down to provide optimal performance and cost-efficiency.

SageMaker synchronous inference endpoints are typically scaled using the InvocationsPerInstance metric, which helps determine event triggers based on real-time demands. However, for SageMaker asynchronous endpoints, this metric isn’t available due to their asynchronous nature.

We encountered challenges with alternative metrics such as ApproximateBacklogSizePerInstance because they didn’t meet our real-time requirements. The inherent delay in these metrics resulted in unacceptable latency in our scaling processes.

Consequently, we sought a custom metric that could more accurately reflect the real-time load on our SageMaker instances.

Amazon CloudWatch custom metrics provide a powerful tool for monitoring and managing your applications and services in the AWS Cloud.

We had previously established a range of custom metrics to monitor various aspects of our infrastructure, including a particularly crucial one for tracking cache misses during image generation. Due to the nature of asynchronous endpoints, which don’t provide the InvocationsPerInstance metric, this custom cache miss metric became essential. It enabled us to gauge the number of requests contributing to the size of the endpoint queue. With this insight into the number of requests, one of our senior developers began to explore additional metrics available through CloudWatch to calculate the asynchronous endpoint capacity and utilization rate. We used the following calculations:

• InferenceCapacity = (CPU utilization * 60) / (InferenceTimeInSeconds * InstanceGPUCount)
• Number of inference requests = (served from cache + cache misses)
• Usage rate = (number of requests) / (InferenceCapacity)​

The calculations included the following variables:

• CPU utilization – Represents the average CPU utilization percentage of the SageMaker instances (CPUUtilization CloudWatch metric). It provides a snapshot of how much CPU resources are currently being used by the instances.
• InferenceCapacity – The total number of inference tasks that the system can process per minute, calculated based on the average CPU utilization and scaled by the number of GPUs available (inf2.48xlarge has 12 GPUs). This metric provides an estimate of the system’s throughput capability per minute.
  • Multiply by 60 / Divide by InferenceTimeInSeconds – This step effectively adjusts the CPUUtilization metric to reflect how it translates into jobs per minute, assuming each job takes 10 seconds. Therefore, (CPU utilization * 60) / 10 represents the theoretical maximum number of jobs that can be processed in one minute based on current or typical CPU utilization.
  • Multiply by 12 – Because the inf2.48xlarge instance has 12 GPUs, this multiplication provides a total capacity in terms of how many jobs all GPUs can handle collectively in 1 minute.
• Number of inference requests (served from cache + cache misses) – We monitor the total number of inference requests processed, distinguishing between those served from cache and those requiring real-time processing due to cache misses. This helps us gauge the overall workload.
• Usage rate (number of inference requests) / (InferenceCapacity)​ – This formula determines the rate of resource usage by comparing the number of operations that invoke new tasks (number of requests) to the total inference capacity (InferenceCapacity).

A higher InferenceCapacity value suggests that we have either scaled up our resources or that our instances are under-utilized. Conversely, a lower capacity value could indicate that we’re reaching our capacity limits and might need to scale out to maintain performance.

Our custom usage rate metric quantifies the usage rate of available SageMaker instance capacity. It’s a composite measure that factors in both the image generation tasks that weren’t served from cache and those that resulted in a cache miss, relative to the total capacity metric. The usage rate is intended to provide insights into how much of the total provisioned SageMaker instance capacity is actively being used for image generation operations. It serves as a key indicator of operational efficiency and helps identify the workload’s operational demands.

We then used the usage rate metric as our auto scaling trigger metric. The use of this trigger in our auto scaling policy made sure SageMaker instances were neither over-provisioned nor under-provisioned. A high value for usage rate might indicate the need to scale up resources to maintain performance. A low value, on the other hand, could signal under-utilization, indicating a potential for cost optimization by scaling down resources.

We applied our custom metrics as triggers for a scaling policy:

CustomizedMetricSpecification = {
    "Metrics": [
        {
            "Id": "m1",
            "MetricStat": {
                "Metric": {
                    "MetricName": "CPUUtilization",
                    "Namespace": "/aws/sagemaker/Endpoints",
                    "Dimensions": [
                        { "Name": "EndpointName", "Value": endpoint_name },
                        { "Name": "VariantName", "Value": "AllTraffic" },
                    ]
                },
                "Stat": "SampleCount"
            },
            "ReturnData": False
        },
        {
            "Id": "m2",
            "MetricStat": {
                "Metric": {
                    "MetricName": " NumberOfInferenceRequests ",
                    "Namespace": "ImageGenAPI",
                    "Dimensions": [
                        { "Name": "service", "Value": "ImageGenerator" },
                        { "Name": "executionEnv", "Value": "AWS_Lambda_nodejs18.x" },
                        { "Name": "region", "Value": "us-west-2" },
                    ]
                },
                "Stat": "SampleCount"
            },
            "ReturnData": False
        },
        {
            "Label": "utilization rate",
            "Id": "e1",
            "Expression": "IF(m1 != 0, m2 / (m1 * 60 / 10 * 12))",
            "ReturnData": True
        }
    ]
}

aas_client.put_scaling_policy(
    PolicyName=endpoint_name,
    PolicyType="TargetTrackingScaling",
    ServiceNamespace=service_namespace,
    ResourceId=resource_id,
    ScalableDimension=scalable_dimension,
    TargetTrackingScalingPolicyConfiguration={
        "CustomizedMetricSpecification": CustomizedMetricSpecification,
        "TargetValue":0.75,
        "ScaleOutCooldown": 60,
        "ScaleInCooldown": 120,
        "DisableScaleIn": False,
    }
)

Deployment on AWS Inferentia2 chips

The integration of AWS Inferentia2 chips into our SageMaker inference endpoints not only resulted in a four-times increase in inference performance for our finely-tuned Stable Diffusion XL model, but also significantly enhanced cost-efficiency. Specifically, SageMaker instances powered by these chips reduced our deployment costs by 60% compared to other comparable instances on AWS. This substantial reduction in cost, coupled with improved performance, underscores the value of using AWS Inferentia2 for intensive computational tasks such as real-time diffusion AI image generation.

Given the importance of swift response times for our specific use case, we established an acceptance criterion of single digit second latency.

SageMaker instances equipped with AWS Inferentia2 chips successfully optimized our infrastructure to deliver image generation in just 9.7 seconds. This enhancement not only met our performance requirements at a low cost, but also provided a seamless and engaging user experience owing to the high availability of Inferentia2 chips.

The effort to integrate with the Neuron SDK also proved highly beneficial. The optimized model not only met our performance criteria, but also enhanced the overall efficiency of our inference processes.

Results and benefits

The implementation of SageMaker asynchronous inference endpoints significantly enhanced our architecture’s ability to handle varying traffic loads and optimize resource utilization, leading to marked improvements in performance and cost-efficiency:

• Inference performance – The AWS Inferentia2 setup processed an average of 27,796 images per instance per hour, giving us 2x improvement in throughput over comparable accelerated compute instances.
• Inference savings – In addition to performance enhancements, the AWS Inferentia2 configurations achieved a 60% reduction in cost per image compared to the original estimation. The cost for processing each image with AWS Inferentia2 was $0.000425. Although the initial requirement to compile models for the AWS Inferentia2 chips introduced an additional time investment, the substantial throughput gains and significant cost reductions justified this effort. For demanding workloads that necessitate high throughput without compromising budget constraints, AWS Inferentia2 instances are certainly worthy of consideration.
• Smoothing out traffic spikes – We effectively smoothed out spikes in traffic to provide continual real-time experience for end-users. As shown in the following figure, the SageMaker asynchronous endpoint auto scaling and managed queue is preventing significant drift from our goal of single digit second latency per image generation.

• Scheduled scaling to manage demand – We can scale up and back down on schedule to cover more predictable traffic demands, reducing inference costs while supplying demand. The following figure illustrates the impact of auto scaling reacting to unexpected demand as well as scaling up and down on a schedule.

Conclusion

In this post, we discussed the potential benefits of applying SageMaker and AWS Inferentia2 chips within a production-ready generative AI application. SageMaker fully managed asynchronous endpoints provide an application time to react to both unexpected and predictable demand in a structured manner, even for high-demand applications such as image-based generative AI. Despite the learning curve involved in compiling the Stable Diffusion XL model to run on AWS Inferentia2 chips, using AWS Inferentia2 allowed us to achieve our demanding low-latency inference requirements, providing an excellent user experience, all while remaining cost-efficient.

To learn more about SageMaker deployment options for your generative AI use cases, refer to the blog series Model hosting patterns in Amazon SageMaker. You can get started with hosting a Stable Diffusion model with SageMaker and AWS Inferentia2 by using the following example.

Discover how Monks serves as a comprehensive digital partner by integrating a wide array of solutions. These encompass media, data, social platforms, studio production, brand strategy, and cutting-edge technology. Through this integration, Monks enables efficient content creation, scalable experiences, and AI-driven data insights, all powered by top-tier industry talent.

About the Authors

Benjamin Moody is a Senior Solutions Architect at Monks. He focuses on designing and managing high-performance, robust, and secure architectures, utilizing a broad range of AWS services. Ben is particularly adept at handling projects with complex requirements, including those involving generative AI at scale. Outside of work, he enjoys snowboarding and traveling.

Karan Jain is a Senior Machine Learning Specialist at AWS, where he leads the worldwide Go-To-Market strategy for Amazon SageMaker Inference. He helps customers accelerate their generative AI and ML journey on AWS by providing guidance on deployment, cost-optimization, and GTM strategy. He has led product, marketing, and business development efforts across industries for over 10 years, and is passionate about mapping complex service features to customer solutions.

Raghu Ramesha is a Senior Gen AI/ML Specialist Solutions Architect with AWS. He focuses on helping enterprise customers build and deploy AI/ ML production workloads to Amazon SageMaker at scale. He specializes in generative AI, machine learning, and computer vision domains, and holds a master’s degree in Computer Science from UT Dallas. In his free time, he enjoys traveling and photography.

Rupinder Grewal is a Senior Gen AI/ML Specialist Solutions Architect with AWS. He currently focuses on model serving and MLOps on SageMaker. Prior to this role, he worked as a Machine Learning Engineer building and hosting models. Outside of work, he enjoys playing tennis and biking on mountain trails.

Parag Srivastava is a Senior Solutions Architect at AWS, where he has been helping customers in successfully applying generative AI to real-life business scenarios. During his professional career, he has been extensively involved in complex digital transformation projects. He is also passionate about building innovative solutions around geospatial aspects of addresses.

Read More

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading artificial intelligence (AI) companies like AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon through a single API, along with a broad set of capabilities to build generative AI applications with security, privacy, and responsible AI.

With Amazon Bedrock, you can experiment with and evaluate top FMs for various use cases. It allows you to privately customize them with your enterprise data using techniques like Retrieval Augmented Generation (RAG), and build agents that run tasks using your enterprise systems and data sources. Knowledge Bases for Amazon Bedrock enables you to aggregate data sources into a repository of information. With knowledge bases, you can effortlessly build an application that takes advantage of RAG.

Accessing up-to-date and comprehensive information from various websites is crucial for many AI applications in order to have accurate and relevant data. Customers using Knowledge Bases for Amazon Bedrock want to extend the capability to crawl and index their public-facing websites. By integrating web crawlers into the knowledge base, you can gather and utilize this web data efficiently. In this post, we explore how to achieve this seamlessly.

Web crawler for knowledge bases

With a web crawler data source in the knowledge base, you can create a generative AI web application for your end-users based on the website data you crawl using either the AWS Management Console or the API. The default crawling behavior of the web connector starts by fetching the provided seed URLs and then traversing all child links within the same top primary domain (TPD) and having the same or deeper URL path.

The current considerations are that the URL can’t require any authentication, it can’t be an IP address for its host, and its scheme has to start with either http:// or https://. Additionally, the web connector will fetch non-HTML supported files such as PDFs, text files, markdown files, and CSVs referenced in the crawled pages regardless of their URL, as long as they aren’t explicitly excluded. If multiple seed URLs are provided, the web connector will crawl a URL if it fits any seed URL’s TPD and path. You can have up to 10 source URLs, which the knowledge base uses to as a starting point to crawl.

However, the web connector doesn’t traverse pages across different domains by default. The default behavior, however, will retrieve supported non-HTML files. This makes sure the crawling process remains within the specified boundaries, maintaining focus and relevance to the targeted data sources.

Understanding the sync scope

When setting up a knowledge base with web crawl functionality, you can choose from different sync types to control which webpages are included. The following table shows the example paths that will be crawled given the source URL for different sync scopes (https://example.com is used for illustration purposes).

You can set the maximum throttling for crawling speed to control the maximum crawl rate. Higher values will reduce the sync time. However, the crawling job will always adhere to the domain’s robots.txt file if one is present, respecting standard robots.txt directives like ‘Allow’, ‘Disallow’, and crawl rate.

You can further refine the scope of URLs to crawl by using inclusion and exclusion filters. These filters are regular expression (regex) patterns applied to each URL. If a URL matches any exclusion filter, it will be ignored. Conversely, if inclusion filters are set, the crawler will only process URLs that match at least one of these filters that are still within the scope. For example, to exclude URLs ending in .pdf, you can use the regex ^.*.pdf$. To include only URLs containing the word “products,” you can use the regex .*products.*.

Solution overview

In the following sections, we walk through the steps to create a knowledge base with a web crawler and test it. We also show how to create a knowledge base with a specific embedding model and an Amazon OpenSearch Service vector collection as a vector database, and discuss how to monitor your web crawler.

Prerequisites

Make sure you have permission to crawl the URLs you intend to use, and adhere to the Amazon Acceptable Use Policy. Also make sure any bot detection features are turned off for those URLs. A web crawler in a knowledge base uses the user-agent bedrockbot when crawling webpages.

Create a knowledge base with a web crawler

Complete the following steps to implement a web crawler in your knowledge base:

1. On the Amazon Bedrock console, in the navigation pane, choose Knowledge bases.
2. Choose Create knowledge base.
3. On the Provide knowledge base details page, set up the following configurations:
   1. Provide a name for your knowledge base.
   2. In the IAM permissions section, select Create and use a new service role.
   3. In the Choose data source section, select Web Crawler as the data source.
   4. Choose Next.
4. On the Configure data source page, set up the following configurations:
   1. Under Source URLs, enter https://www.aboutamazon.com/news/amazon-offices.
   2. For Sync scope, select Host only.
   3. For Include patterns, enter ^https?://www.aboutamazon.com/news/amazon-offices/.*$.
   4. For exclude pattern, enter .*plants.* (we don’t want any post with a URL containing the word “plants”).
   5. For Content chunking and parsing, chose Default.
   6. Choose Next.
5. On the Select embeddings model and configure vector store page, set up the following configurations:
   1. In the Embeddings model section, chose Titan Text Embeddings v2.
   2. For Vector dimensions, enter 1024.
   3. For Vector database, choose Quick create a new vector store.
   4. Choose Next.
6. Review the details and choose Create knowledge base.

In the preceding instructions, the combination of Include patterns and Host only sync scope is used to demonstrate the use of the include pattern for web crawling. The same results can be achieved with the default sync scope, as we learned in the previous section of this post.

You can use the Quick create vector store option when creating the knowledge base to create an Amazon OpenSearch Serverless vector search collection. With this option, a public vector search collection and vector index is set up for you with the required fields and necessary configurations. Additionally, Knowledge Bases for Amazon Bedrock manages the end-to-end ingestion and query workflows.

Test the knowledge base

Let’s go over the steps to test the knowledge base with a web crawler as the data source:

1. On the Amazon Bedrock console, navigate to the knowledge base that you created.
2. Under Data source, select the data source name and choose Sync. It could take several minutes to hours to sync, depending on the size of your data.

3. When the sync job is complete, in the right panel, under Test knowledge base, choose Select model and select the model of your choice.
4. Enter one of the following prompts and observe the response from the model:
   1. How do I tour the Seattle Amazon offices?
   2. Provide me with some information about Amazon’s HQ2.
   3. What is it like in the Amazon’s New York office?

As shown in the following screenshot, citations are returned within the response reference webpages. The value of x-amz-bedrock-kb-source-uri is a webpage link, which helps you verify the response accuracy.

Create a knowledge base using the AWS SDK

This following code uses the AWS SDK for Python (Boto3) to create a knowledge base in Amazon Bedrock with a specific embedding model and OpenSearch Service vector collection as a vector database:

import boto3

client = boto3.client('bedrock-agent')

response = client.create_knowledge_base(
    name='workshop-aoss-knowledge-base',
    roleArn='your-role-arn',
    knowledgeBaseConfiguration={
        'type': 'VECTOR',
        'vectorKnowledgeBaseConfiguration': {
            'embeddingModelArn': 'arn:aws:bedrock:your-region::foundation-model/amazon.titan-embed-text-v2:0'
        }
    },
    storageConfiguration={
        'type': 'OPENSEARCH_SERVERLESS',
        'opensearchServerlessConfiguration': {
            'collectionArn': 'your-opensearch-collection-arn',
            'vectorIndexName': 'blog_index',
            'fieldMapping': {
                'vectorField': 'documentid',
                'textField': 'data',
                'metadataField': 'metadata'
            }
        }
    }
)

The following Python code uses Boto3 to create a web crawler data source for an Amazon Bedrock knowledge base, specifying URL seeds, crawling limits, and inclusion and exclusion filters:

import boto3

client = boto3.client('bedrock-agent', region_name='us-east-1')

knowledge_base_id = 'knowledge-base-id'

response = client.create_data_source(
    knowledgeBaseId=knowledge_base_id,
    name='example',
    description='test description',
    dataSourceConfiguration={
        'type': 'WEB',
        'webConfiguration': {
            'sourceConfiguration': {
                'urlConfiguration': {
                    'seedUrls': [
                        {'url': 'https://example.com/'}
                    ]
                }
            },
            'crawlerConfiguration': {
                'crawlerLimits': {
                    'rateLimit': 300
                },
                'inclusionFilters': [
                    '.*products.*'
                ],
                'exclusionFilters': [
                    '.*.pdf$'
                ],
                'scope': 'HOST_ONLY'
            }
        }
    }
)

Monitoring

You can track the status of an ongoing web crawl in your Amazon CloudWatch logs, which should report the URLs being visited and whether they are successfully retrieved, skipped, or failed. The following screenshot shows the CloudWatch logs for the crawl job.

Clean up

To clean up your resources, complete the following steps:

1. Delete the knowledge base:
   1. On the Amazon Bedrock console, choose Knowledge bases under Orchestration in the navigation pane.
   2. Choose the knowledge base you created.
   3. Take note of the AWS Identity and Access Management (IAM) service role name in the knowledge base overview.
   4. In the Vector database section, take note of the OpenSearch Serverless collection ARN.
   5. Choose Delete, then enter delete to confirm.
2. Delete the vector database:
   1. On the OpenSearch Service console, choose Collections under Serverless in the navigation pane.
   2. Enter the collection ARN you saved in the search bar.
   3. Select the collection and chose Delete.
   4. Enter confirm in the confirmation prompt, then choose Delete.
3. Delete the IAM service role:
   1. On the IAM console, choose Roles in the navigation pane.
   2. Search for the role name you noted earlier.
   3. Select the role and choose Delete.
   4. Enter the role name in the confirmation prompt and delete the role.

Conclusion

In this post, we showcased how Knowledge Bases for Amazon Bedrock now supports the web data source, enabling you to index public webpages. This feature allows you to efficiently crawl and index websites, so your knowledge base includes diverse and relevant information from the web. By taking advantage of the infrastructure of Amazon Bedrock, you can enhance the accuracy and effectiveness of your generative AI applications with up-to-date and comprehensive data.

For pricing information, see Amazon Bedrock pricing. To get started using Knowledge Bases for Amazon Bedrock, refer to Create a knowledge base. For deep-dive technical content, refer to Crawl web pages for your Amazon Bedrock knowledge base. To learn how our Builder communities are using Amazon Bedrock in their solutions, visit our community.aws website.

About the Authors

Hardik Vasa is a Senior Solutions Architect at AWS. He focuses on Generative AI and Serverless technologies, helping customers make the best use of AWS services. Hardik shares his knowledge at various conferences and workshops. In his free time, he enjoys learning about new tech, playing video games, and spending time with his family.

Malini Chatterjee is a Senior Solutions Architect at AWS. She provides guidance to AWS customers on their workloads across a variety of AWS technologies. She brings a breadth of expertise in Data Analytics and Machine Learning. Prior to joining AWS, she was architecting data solutions in financial industries. She is very passionate about semi-classical dancing and performs in community events. She loves traveling and spending time with her family.

Read More

Intuit is committed to providing its customers innovative solutions that simplify complex financial processes. Tax filing can be a challenge, with its ever-changing regulations and intricate nuances. That’s why the company empowers millions of individuals and small businesses to comprehend tax-related information effortlessly and file with full confidence that their taxes are done right.

For the 2024 tax season, Intuit set out to raise the bar with generative AI, using Anthropic’s advanced language model Claude in Amazon Bedrock—underpinned by Intuit’s proprietary tax engine—to provide individual tax filers with simple-to-understand contextual explanations of tax calculations, backed by real-time accuracy checks.

In this blog post, we discuss the journey of developing a solution that benefited millions of TurboTax customers in 2024.

The challenge

Taxes, with their complicated regulations and nuances, can be a labyrinth for even the most seasoned. The tax code includes 15,000+ federal tax forms and state tax forms for individual and business tax filers in the U.S. It is estimated that Americans spend 8.9 billion hours every year doing their taxes.

To streamline and simplify the tax filing experiences, Intuit’s AI/GenAI-powered TurboTax products guide consumers through the process. One challenge is to explain complex calculations in a simple-to-understand manner so taxpayers can confidently file their taxes —and seamlessly connect to a human expert whenever needed. According to Nhung Ho, vice president of AI at Intuit, “With Intuit Assist for TurboTax, we wanted to answer every customer’s question about how they arrived at their final tax outcome, and we had to do it in clear, concise language, so they have peace of mind before they file.”

The solution

Applying its years of domain expertise, robust data set and proprietary tax knowledge engine, Intuit worked closely with Anthropic and Amazon Web Services to further boost filer confidence by integrating Claude via Amazon Bedrock into its AI financial assistant, Intuit Assist for TurboTax. During federal tax reviews where customers see a summary of their return, the combined work of Intuit, Anthropic and AWS provides simple explanations of tax calculations. Altogether, helping users better understand how their tax result is calculated will help them feel assured their taxes were filed correctly. The following video shows examples of tax explanations.

Implementing Claude in Amazon Bedrock: a collaborative effort

In June 2023, Intuit announced its proprietary generative AI operating system (GenOS), which runs on AWS infrastructure and empowers the company’s developers to design, build, and deploy breakthrough generative AI experiences. GenOS serves as the primary paved path for rolling out generative AI applications or capabilities in production across the company.

Last fall, Intuit began experimenting with Anthropic’s Claude via Amazon Bedrock.

“After a successful partnership with Amazon SageMaker for its ML capabilities, Intuit looked forward to working with Amazon Bedrock as a managed service to simplify the deployment and management of LLMs,” explained Nhung.

Each year, tax filing is a seasonal process between January 1 and October 15, so the ability to scale rapidly to help meet the needs of millions of Intuit customers during this period was a critical success factor for Intuit’s tax explanations use case with Anthropic Claude in Amazon Bedrock.

“Amazon Bedrock offered Intuit the latency, scalability, and reliability to introduce AI-powered tax explanations to its customers,” Nhung added. “This allowed Intuit to deliver valuable generative AI experiences to its users.”

The company took advantage of AWS elasticity to acquire resources as they needed them, and to release resources when no longer needed. Provisioned throughput for Amazon Bedrock enabled Intuit to achieve the scalability and latency needed to serve millions of customers, beginning in January 2024. Intuit also implemented a multi-region setup to provide resiliency needed for such a critical application.

Additionally, a private connection between TurboTax Virtual Private Cloud (VPC) and Amazon Bedrock made sure that user data was appropriately protected.

“Intuit takes great pains to protect user data with our anti-fraud technology. It is important that user data remain secure. Anthropic’s Claude LLM, managed by Amazon Bedrock, provides that capability.” Nhung explained.

Conclusion

By using Amazon Bedrock to integrate Anthropic’s Claude into its tax preparation software, Intuit expanded the following benefits:

• Simplified Tax Explanations: By demystifying tax complexities, Intuit instilled confidence in users, empowering them to navigate the tax filing process with greater ease and assurance.

• Simplified Management: A simplified management experience of Anthropic’s Claude with Bedrock made it simple for Intuit to scale securely.

For the 2024 tax season, Intuit’s innovative use of Anthropic’s Claude in Amazon Bedrock is helping demystify the complexities of tax filing. By harnessing the power of advanced language models, the company is redefining the way people understand and engage with tax-related information. Through personalized explanations, tailored guidance, and a commitment to continuous improvement, Intuit is paving the way for a “done for you” future, where the hard work of tax preparation is done on its customers’ behalf, with a seamless path to human tax and bookkeeping experts whenever needed.

As the company moves forward, it remains dedicated to using cutting-edge generative AI technologies to enhance its solutions and provide its customers with the tools they need to achieve financial success. The successful integration of Amazon Bedrock in the tax domain has opened up new opportunities for Intuit to leverage advanced language models in other areas of financial management, solidifying its position as a trailblazer in fintech.

About the Author

Shivanshu Upadhyay is a Principal Solutions Architect in the AWS Industries group. In this role, he helps the most advanced adopters of AWS transform their industry by effectively using data and AI.

Read More