Skip to main content

Minimal Bedrock Document Processing

Code Documentation Construct

The absolute minimal document processing system using AppMod Catalog Blueprints. This example shows how to deploy a complete AI-powered document processing pipeline with zero configuration.

What This Example Does​

Deploy a complete document processing system with just one construct:

  • Document Upload: S3 bucket automatically configured for document uploads
  • AI Classification: Automatically identifies document types using Amazon Bedrock
  • Data Extraction: Extracts key information from documents using Claude 3.5 Sonnet
  • Workflow Orchestration: Step Functions manages the entire processing pipeline
  • Results Storage: DynamoDB stores processing results and metadata

Architecture​

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   S3 Bucket     │    │   Step Functions│    │   Amazon        │
│   Document      │───▶│   Workflow      │───▶│   Bedrock       │
│   Upload        │    │                 │    │   (Claude 3.5)  │
└─────────────────┘    └─────────────────┘    └─────────────────┘
                                │                        │
                                ▼                        ▼
                       ┌─────────────────┐    ┌─────────────────┐
                       │   DynamoDB      │    │   CloudWatch    │
                       │   Results       │    │   Logs          │
                       └─────────────────┘    └─────────────────┘

The Code​

The entire system is deployed with just a few lines:

// Create S3 bucket for document storage
const bucket = new Bucket(this, 'DocumentBucket', {
  encryption: BucketEncryption.KMS,
});

// Deploy complete document processing!
new BedrockDocumentProcessing(this, 'MinimalDocProcessor', {
  ingressAdapter: new QueuedS3Adapter({ bucket }),
  // Enable cross-region inference for Claude Sonnet 4
  classificationBedrockModel: {
    useCrossRegionInference: true,
  },
  processingBedrockModel: {
    useCrossRegionInference: true,
  },
});

This creates:

  • S3 bucket with proper event notifications
  • Step Functions workflow for processing orchestration
  • Lambda functions for classification and extraction
  • DynamoDB table for results storage
  • IAM roles with least-privilege permissions
  • CloudWatch logs for monitoring

Quick Start​

Prerequisites​

  • AWS CLI configured with appropriate permissions
  • CDK CLI installed: npm install -g aws-cdk
  • Node.js 18+
  • Docker Desktop running (required for Python Lambda bundling)
  • Amazon Bedrock model access (Claude 3.5 Sonnet)

Deploy​

# Install dependencies
npm install

# Deploy the stack
npm run deploy

Deployment time: ~5 minutes

Outputs: After deployment, you'll see:

  • DocumentBucketName: S3 bucket name for uploading documents
  • StateMachineArn: Step Functions workflow ARN for monitoring
  • ProcessingTableName: DynamoDB table name for results

Test Document Processing​

After deployment, upload a document to trigger processing:

Option 1: Using the upload script (recommended)

# Make the script executable
chmod +x upload-document.sh

# Upload a test document
./upload-document.sh sample-files/sample-invoice.jpg

Option 2: Using AWS CLI directly

# Get bucket name from deployment outputs
BUCKET_NAME=$(aws cloudformation describe-stacks \
    --stack-name MinimalBedrockDocProcessingStack \
    --query 'Stacks[0].Outputs[?OutputKey==`DocumentBucketName`].OutputValue' \
    --output text)

# Upload document to raw/ prefix to trigger processing
aws s3 cp sample-files/sample-invoice.jpg s3://$BUCKET_NAME/raw/

View Results:

  • Step Functions Console: Search for "MinimalBedrockDocProcessingStack" to view workflow executions
  • DynamoDB Console: Open the processing table (from outputs) to see extracted data
  • CloudWatch Logs: Check /aws/lambda/ log groups for detailed processing logs

Default Capabilities​

Out of the box, this system can process:

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

What's Included​

The BedrockDocumentProcessing construct provides:

  • Automatic retries for transient failures
  • Error handling with proper error states
  • Cost optimization with appropriate resource sizing
  • Security with encryption at rest and in transit
  • Monitoring with CloudWatch metrics and logs
  • Scalability with serverless auto-scaling

Troubleshooting​

Docker Not Running​

Error: Cannot connect to the Docker daemon

Solution: Start Docker Desktop on your machine. The construct uses Python Lambda functions that require Docker for bundling dependencies during CDK synthesis.

Deployment Issues​

  • Bedrock access denied: Verify you have model access in the Bedrock console (us-east-1 or your deployment region)
  • No processing triggered: Ensure documents are uploaded to the S3 bucket (check CloudFormation outputs for bucket name)
  • Processing timeout: Check document size - very large documents may need timeout adjustments

Monitoring​

  • Step Functions Console: View workflow execution progress and any errors
  • CloudWatch Logs: Check Lambda function logs for detailed processing information
  • DynamoDB Console: Query the table to see processing results

Next Steps​

This example demonstrates the power of AppMod Blueprints defaults. For more advanced features, see:

Cleanup​

npm run destroy