Skip to main content

Bedrock Document Processing

Overview​

This example demonstrates AI-powered document processing using the BedrockDocumentProcessing construct with Amazon Bedrock foundation models for intelligent document classification and data extraction.

Use Case: Invoice processing, document digitization, content extraction

What This Example Does​

  • Document Classification: Automatically identifies document types (invoices, receipts, identity documents, etc.)
  • Information Extraction: Extracts key entities and data from documents using AI
  • Workflow Orchestration: Uses Step Functions to manage the processing pipeline
  • Metadata Storage: Tracks processing status and results in DynamoDB

Architecture​

Key Features​

  • AI-Powered Processing: Uses Claude 3.7 Sonnet for classification and extraction
  • Automatic Triggering: S3 uploads automatically start processing
  • Structured Results: JSON output with extracted entities and metadata
  • Error Handling: Failed documents moved to failed/ prefix with error details
  • Observability: CloudWatch logs and Step Functions monitoring

Deployment​

Prerequisites​

  • AWS CLI configured with appropriate permissions
  • CDK CLI installed: npm install -g aws-cdk
  • Node.js 18+
  • Amazon Bedrock model access (Claude 3.7 Sonnet)

Deploy Steps​

# Install dependencies
npm install

# Build the project
npm run build

# Deploy the stack
npx cdk deploy --require-approval never

Usage Example​

1. Upload Documents​

Upload documents to the S3 bucket's raw/ prefix:

# Get bucket name from CDK output
aws s3 cp sample-invoice.pdf s3://YOUR-BUCKET-NAME/raw/

2. Monitor Processing​

  • Step Functions Console: View workflow execution progress
  • CloudWatch Logs: Check Lambda function logs for detailed processing
  • DynamoDB: Query metadata table for processing results

3. Processing Results​

The workflow produces structured JSON output:

{
  "documentId": "sample-invoice-1756401373049",
  "classificationResult": {
    "documentClassification": "INVOICE"
  },
  "processingResult": {
    "result": {
      "entities": [
        {
          "type": "INVOICE_NUMBER",
          "value": "INV-2024-001"
        },
        {
          "type": "TOTAL_AMOUNT",
          "value": "$1,250.00"
        },
        {
          "type": "DUE_DATE",
          "value": "2024-02-15"
        }
      ]
    }
  }
}

Supported Document Types​

  • Invoices: Invoice number, amounts, dates, vendor information
  • Receipts: Purchase details, merchant info, totals
  • Identity Documents: Names, addresses, ID numbers
  • Forms: Field extraction based on document structure

Configuration Options​

new BedrockDocumentProcessing(this, 'DocProcessing', {
  // Custom models
  classificationModelId: FoundationModelIdentifier.ANTHROPIC_CLAUDE_3_HAIKU_20240307_V1_0,
  processingModelId: FoundationModelIdentifier.ANTHROPIC_CLAUDE_3_SONNET_20240229_V1_0,
  
  // Custom prompts
  classificationPrompt: "Your custom classification prompt...",
  processingPrompt: "Your custom extraction prompt...",
  
  // Enable observability
  enableObservability: true
});

Troubleshooting​

Common Issues:

  1. No processing triggered: Ensure documents are uploaded to raw/ prefix
  2. Bedrock access denied: Verify model access in Bedrock console
  3. Processing timeout: Check document size and complexity

Monitoring:

  • Step Functions execution history
  • CloudWatch metrics for Lambda functions
  • DynamoDB items for processing status

Cleanup​

npx cdk destroy

Note: This will delete all resources including the S3 bucket and any uploaded documents.