Claims Review Bedrock Agent Stack Documentation

Overview

The claims-review is an AWS CDK stack that sets up an Amazon Bedrock agentic architecture to automate the processing and review of medical insurance claim forms using Amazon Bedrock Data Automation.

Prerequisites

Python 3.10 or higher
AWS CLI configured with appropriate credentials
Node.js and npm (for AWS CDK CLI)
AWS CDK CLI installed (npm install -g aws-cdk)

Select a Foundation Model to use with Bedrock Agent.

Before deploying the stack, you need to choose a foundation model (FM) that the agent invokes to interpret user input and subsequent prompts in its orchestration process. The agent also invokes the FM to generate responses and follow-up steps in its process. The following sections provide information on selecting a foundation model and enabling access to the model.

For more details see How Amazon Bedrock Agents works.

Choosing a Model or an Inference Profile

When choosing a model please follow the model provider acceptable end user policy

Model Support By Feature and By Region

Foundation models differ in the Amazon Bedrock features that they support. See Model support by feature
Amazon Bedrock foundation models differ in their regional support. See Model support by AWS Region in Amazon Bedrock

[!Note] Some models are accessible in some Regions only through cross-region inference. To learn more about cross-region inference, see Increase throughput with cross-region inference and Supported Regions and models for inference profiles.

Access Amazon Bedrock foundation models

Before you can use a foundation model in Amazon Bedrock, you must request access to it.

See Add or remove access to Amazon Bedrock foundation model
If choosing one of Amazon Titan Text models ensure the model Supported use cases include Agents Support

Using Inference Profiles

You can use a cross region inference profile in place of a foundation model to route requests to multiple Regions. When using inference profiles, make sure you've requested access to the models and the regions defined in the inference profiles that you want to use. For example, to gain access to make calls to the US Anthropic Claude 3 Haiku inference profile from the US West (Oregon) Region, do the following:

Sign into the AWS Management Console in the US East (N. Virginia) Region and request model access to Anthropic Claude 3 Haiku by following the steps at Access Amazon Bedrock foundation models.
Change to the US West (Oregon) Region and request model access to Anthropic Claude 3 Haiku.

Important

Take a note of the foundation model id or the inference profile id. You would need to use the model id with the cdk deploy command

Deployment Steps

Note

If you’re continuing this part from Installation part 1, you can skip step 1-4

Create a new directory, navigate to that directory in a terminal and clone the GitHub repository:

git clone https://github.com/aws-solutions-library-samples/guidance-for-multimodal-data-processing-using-amazon-bedrock-data-automation.git

Change to the deployment directory for the guidance repository

cd guidance-for-multimodal-data-processing-using-amazon-bedrock-data-automation/deployment

Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate

Install required dependencies:
```
pip install -r requirements.txt
```
Bootstrap AWS CDK (first-time only):
```
cdk bootstrap
```

Go to the layer directory and install lambda layer dependencies into the python subdirectory:

cd lambda/claims_review/layer/
pip install -r requirements.txt --target python
cd ../../..

Deploy the stack:

[!Note] You would need a model id or an inference profile id. See Select a Foundation Model

Using a foundation model
```
cdk deploy claims-review  --context foundation_model_id=<<your_chosen_model_id>>
```
Using an inference profile
```
cdk deploy claims-review  --context inference_profile_id=<<your_chosen_inference_profile_id>>
```
[!Important] You must provide one of foundation model id or inference profile id, but not both

To protect you against unintended changes that affect your security posture, the CDK CLI prompts you to approve security-related changes before deploying them. When prompted, review the changes and Enter y for Do you wish to deploy these changes (y/n)? if you intend to proceed.

Alternatively, in one command

Using a foundation model
```
cdk deploy claims-review  --context foundation_model_id=<<your_chosen_model_id>> --require-approval never 
```
Using an inference profile
```
cdk deploy claims-review  --context inference_profile_id=<<your_chosen_inference_profile_id>> --require-approval never 
```
Wait for the stack deploy to complete. This may take a few minutes.

Deployment Validation

To validate that your AWS CloudFormation stack

Log in to your personal AWS account in the AWS Console.
Navigate to the to the AWS CloudFormation Console by searching for AWS CloudFormation in the search bar at the top of the AWS Console page and Click on CloudFormation in the results.
From the AWS CloudFormation in the Stacks list, look for a stack with Stack name claims-review. Validate that the status shows 'CREATE_COMPLETE'

Alternatively, you can use the AWS CLI

Check the stack status using the AWS CloudFormation service

 aws cloudformation describe-stacks --stack-name claims-review --query 'Stacks[0].StackStatus' --output text

A successful initial deployment should show a CREATE_COMPLETE status and a successful subsequent deployment should show UPDATE_COMPLETE status

Running the Guidance

See the guide here for steps to run the claims review application

Security

The stack enforces the following security measures:

The Bedrock Agent's Resource Role has the minimum permissions required to access the Foundation Model and invoke the Lambda function.
The Lambda function's execution role has the basic execution permissions.
The Lambda function is granted permission to be invoked by the Bedrock Agent.

Troubleshooting

Deployment Issues:

General

Verify AWS credentials are configured correctly
Ensure CDK is bootstrapped in your account/region
Check the CloudFormation console for detailed error messages

`--app is required either in command-line, in cdk.json or in ~/.cdk.json`

Ensure you're in the right directory when running cdk deploy. see Step 1

Runtime Issues:

Check CloudWatch Logs for Lambda function errors
Verify the IAM permissions are correct
Ensure the Bedrock Agent and Action Group are configured correctly

Common errors you might encounter:

"Resource not found": Ensure the required resources exist and the permissions are correct.
"Access denied": Check the IAM roles and policies.
"Invalid handler": Verify the Lambda function handler name.
"Access denied when calling Bedrock": Verify Bedrock Model is available in the region and Model Access has been granted

Development

Key Stack Resources

The stack sets up the following key AWS resources:

Important

The resource names or name prefixes would change if the parameters in cdk.json are modified from their default values before creating the stack

Claims Submission Bucket - Used to store incoming CMS 1500 claim forms
- bucket name prefix claims-review-bdaclaimsreviewbucket
EventBridge Rule - To capture the event when a new claims form is submitted and stored in the Claims Submission Bucket
- Rule name prefix claims-review-bdaonclaimsubmission
Invoke Data Automation Lambda Function - A Lambda function to trigger Bedrock Data Automation job that gathers insights from the claims form.
- Function Name Prefix claims-review-bdainvokedataautomation
EventBridge Rule - To capture the event when a the BDA insight job is completed. The rules has a target Lambda function that then triggers Bedrock Agent
- Rule name prefix claims-review-bdadataautomationcompleted
Invoke Claim Verification Lambda Function - A Lambda function to trigger Bedrock Agent to a prompt to start claim verification steps.
- Function Name Prefix claims-review-bdainvokeverification
Bedrock Agent to review claims
- Agent Nae claims-review-agent
Bedrock Knowledge Base to store Claims Evidence of Coverage (EoC) documents for the various Insurance plans
- Knowledge Base Name claims-eoc-kb
OpenSearch Serverless Collection/Index to serve as vector store for the EoC documents embeddings
- Collection Name claims-vector-store
- Index Name claims_eoc_index
Custom resource with Lambda function to create Opensearch vector index
- function name prefix claims-review-vectorstorecreatevectorindex
Aurora PostgreSQL Serverless Database
- cluster name prefix claims-review-auroraaurorapostgrescluster
- database name claimdatabase
Custom resource with Lambda function to create initial claims database schema and sample data
- function name prefix claims-review-auroraSchemaExecutor
S3 Bucket to store Claims EoC documents
- bucket name prefix claims-review-claimseockbclaimseockbdatasource
EventBridge rule to match event when Claim EoC documents are added/updated. The rule triggers the lambda function to start ingestion job
- rule name prefix claims-review-claimseockbons3objectcreateupdaterule
Lambda Function to trigger datasource Sync for Claims EoC Knowledge Base
- function name prefix claims-review-datasourcesynclambdafunction

Project Structure

Click for Project Structure

guidance-for-multimodal-data-processing-using-amazon-bedrock-data-automation/
├── assets/
│   ├── data/
│   │   ├── claims_review/
│   │   │   ├── cms_1500/
│   │   │   │   ├── sample1_cms-1500-P.pdf
│   │   │   │   ├── sample2_cms-1500-P.pdf
│   │   │   │   └── sample3_cms-1500-P.pdf
│   │   │   └── eoc/
│   │   │       ├── Evidence_of_Coverage_-_FakeHealth_Plus.pdf
│   │   │       ├── Evidence_of_Coverage_-_FakeHealth_Premium.pdf
│   │   │       └── Evidence_of_Coverage_-_FakeHealth_Standard.pdf
├── claims-cli.bat
├── claims-cli.sh
├── deployment/
│   ├── app.py
│   ├── cdk.json
│   ├── docs/
│   │   ├── b_claims_review_01_deploy.md
│   │   └── b_claims_review_02_run_flow.md
│   ├── lambda/
│   │   ├── claims_review/
│   │   │   ├── blueprint_creation/
│   │   │   │   └── index.py
│   │   │   ├── claims_review_agent_actions/
│   │   │   │   └── index.py
│   │   │   ├── create_vector_index/
│   │   │   │   └── index.py
│   │   │   ├── datasource_sync/
│   │   │   │   └── index.py
│   │   │   ├── invoke_data_automation/
│   │   │   │   ├── bda_wrapper.py
│   │   │   │   └── index.py
│   │   │   ├── invoke_verification/
│   │   │   │   ├── bedrock_agent_runtime_wrapper.py
│   │   │   │   └── index.py
│   │   │   ├── layer/
│   │   │   │   └── requirements.txt
│   │   │   └── manage_schema/
│   │   │       └── index.py
│   │   └── lending_flow/
│   │       ├── documents_post_processor/
│   │       │   └── index.py
│   │       ├── documents_processor/
│   │       │   └── index.py
│   │       ├── samples_post_processor/
│   │       │   └── index.py
│   │       └── samples_processor/
│   │           └── index.py
│   ├── requirements.txt
│   └── stacks/
│       ├── claims_review_stack/
│       │   ├── agent.py
│       │   ├── aurora_postgres.py
│       │   ├── document_automation.py
│       │   ├── knowledge_base.py
│       │   ├── prompts/
│       │   │   └── claims_review_agent.py
│       │   ├── schemas/
│       │   │   ├── claims_review_openapi.json
│       │   │   ├── create_database_schema.sql
│       │   │   └── delete_database_schema.sql
│       │   └── vector_store.py
├── README.md
├── source/
│   └── claims_review_app/
│       └── claims-cli.py
├── source.bat

Customize the Stack

Customize Stack Parameters

The stack uses context values in cdk.json file to store default parameters used for stack creation. To use your own values, you can either:

Modify the values for the specific keys in cdk.json, or;
use --context parameter with the cdk deploy command to override the values, for example cdk deploy claims-review --context blueprint_name=yyyyy

Customize the Claims Review Bedrock Agent prompt

The prompt instruction used to create the agent is in the deployment/stacks/claims_review_stack/prompts/claims_review_agent.py. To Customize the agent instruction:

Update the text set to variable claims_review_agent_instruction
save the changes
deploy the stack again using instructions in 6. Deploy the stack

Manage Action Group schema

The action group API schema used for the agent action group is in deployment/stacks/claims_review_stack/schemas/claims_review_openapi.json
The Lambda functions backing the action group APIs are in deployment/lambda/claims_review/claims_review_agent_actions/index.py.
The Database schema for the claim database (Aurora Postgres Serverless) is in deployment/stacks/claims_review_stack/schemas/create_database_schema.sql

To Customize the action group API schema:

Update the api schema in the file using OpenAPI 3.0 format
If necessary, make changes to the action group labmda function.
Make necessary changes to the database schema
Save all the changes
Deploy the stack again using instructions in 6. Deploy the stack

Important

In case of breaking/incompatible changes to the database schema, it might be neccesary to delete and redeploy the stack. Follow the steps in Cleanup and 6. Deploy the stack

Update stack resources

The claims-review stack has the following main source code files associated with it -

deployment/stacks/claims_review_stack/agent.py - The top-level claims-review stack that creates bedrock agent resources and uses custom construct to create and configure other related resources include the bedrock knowledge base, vector store, the aurora databaseand BDA automation.
deployment/stacks/claims_review_stack/vector_store.py - Vector store resources include opensearch serverless collection and a Custom resource lambda function to create the vector index for the claims EoC knowledge base
deployment/stacks/claims_review_stack/knowledge_base.py - A construct encapsulating resources required for the bedrock knowledge base including datasource buckets, datasources, the knowledge base itself and KB logging configuration
deployment/stacks/claims_review_stack/document_automation.py - A construct encapsulating resources related to bedrock document automation, include the input/output s3 buckets, lambda function to invoke bda for insights
deployment/stacks/claims_review_stack/aurora_postgres.py - Aurora Postgres database resources for the claims review database including, aurora cluster, database and initial schema creation
deployment/lambda/claims_review - Function code for Lambda functions used in the claims-review stack. The main functions are -
- deployment/lambda/claims_review/claims_review_agent_actions - Bedrock Agent Action group API functions
- deployment/lambda/claims_review/create_vector_index - Function to create the vector index as part of CDK deployment
- deployment/lambda/claims_review/datasource_sync - Function to trigger Bedrock Knowledge Base datasource sync as new documents are uploaded to the backing datasource s3 bucket
- deployment/lambda/claims_review/invoke_data_automation - function to trigger the BDA insight job after a new claim form is submitted
- deployment/lambda/claims_review/invoke_verification - function to trigger claims review bedrock agent after the BDA insight job is completed and claim form data extracted output is available.
- deployment/lambda/claims_review/manage_schema - function to create/update Aurora database schema as part of CDK deployment

To update stack resources -

Update the associated stack or construct file.
Update the Lambda function code in the respective directories, if necessary
Run cdk diff claims-review to review the changes.
Deploy the changes with cdk deploy claims-review.

Cleanup

Ensure that the S3 Buckets for claims submission, claims review and claims Eoc are empty. Save any documents that you would like to retain in an alternate location.

Change to the deployment directory for the guidance repository

cd guidance-for-multimodal-data-processing-using-amazon-bedrock-data-automation

Destroy the stack
```
cdk destroy claims-review
```

Contributing

Create a new branch for features.
Update the documentation as needed.
Test the changes thoroughly.
Submit a pull request.

Useful Links

AWS CDK Python Reference
AWS Lambda Developer Guide
Amazon Bedrock Developer Guide

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

b_claims_review_01_deploy.md

b_claims_review_01_deploy.md

Claims Review Bedrock Agent Stack Documentation

Overview

Prerequisites

Select a Foundation Model to use with Bedrock Agent.

Choosing a Model or an Inference Profile

Model Support By Feature and By Region

Access Amazon Bedrock foundation models

Using Inference Profiles

Deployment Steps

Deployment Validation

Running the Guidance

Security

Troubleshooting

Deployment Issues:

General

`--app is required either in command-line, in cdk.json or in ~/.cdk.json`

Runtime Issues:

Development

Key Stack Resources

Project Structure

Customize the Stack

Customize Stack Parameters

Customize the Claims Review Bedrock Agent prompt

Manage Action Group schema

Update stack resources

Cleanup

Contributing

Useful Links

Files

b_claims_review_01_deploy.md

Latest commit

History

b_claims_review_01_deploy.md

File metadata and controls

Claims Review Bedrock Agent Stack Documentation

Overview

Prerequisites

Select a Foundation Model to use with Bedrock Agent.

Choosing a Model or an Inference Profile

Model Support By Feature and By Region

Access Amazon Bedrock foundation models

Using Inference Profiles

Deployment Steps

Deployment Validation

Running the Guidance

Security

Troubleshooting

Deployment Issues:

General

--app is required either in command-line, in cdk.json or in ~/.cdk.json

Runtime Issues:

Development

Key Stack Resources

Project Structure

Customize the Stack

Customize Stack Parameters

Customize the Claims Review Bedrock Agent prompt

Manage Action Group schema

Update stack resources

Cleanup

Contributing

Useful Links

`--app is required either in command-line, in cdk.json or in ~/.cdk.json`