AI Invoice Parser

An intelligent invoice processing system that receives PDFs via email, extracts structured data using AWS Bedrock Nova AI model, and returns parsed JSON results automatically.

🌟 Features

• Email Integration: Receive invoices via invoice-bot@katechat.tech
• Multi-language Support: Parse invoices in English and German
• AI-Powered Extraction: Uses AWS Bedrock Nova multimodal model with direct PDF input
• Automatic Processing: Fully automated from email receipt to response
• Structured Output: Returns clean JSON with invoice details and line items
• Local Testing: Test parser locally with sample invoices

📁 Project Structure

ai-order-parser/
├── data/                    # Sample invoice PDFs for testing
│   ├── 2025-30240870.pdf
│   ├── 42507956.pdf
│   ├── KR318586338013.pdf
│   ├── Rechnung-38110.pdf
│   └── Rechnung_HDK-054031.pdf
├── processor/               # Lambda function code
│   ├── lambda_function.py   # Main Lambda handler
│   ├── pdf_parser.py        # PDF parsing with Bedrock Nova
│   ├── email_processor.py   # SES email handling
│   ├── test_local.py        # Local testing script
│   ├── requirements.txt     # Python dependencies
│   └── README.md           # Processor documentation
├── infrastructure/          # Terraform IaC
│   ├── main.tf             # Main Terraform configuration
│   ├── variables.tf        # Variable definitions
│   ├── lambda.tf           # Lambda function resources
│   ├── ses.tf              # SES configuration
│   ├── s3.tf               # S3 bucket for emails
│   ├── iam.tf              # IAM roles and policies
│   ├── outputs.tf          # Terraform outputs
│   └── README.md           # Infrastructure documentation
├── deploy.sh               # Deployment script
└── README.md              # This file

🚀 Quick Start

Prerequisites

1. AWS CLI configured with appropriate credentials
2. Terraform >= 1.0 installed
3. Python 3.11+ with pip
4. Domain katechat.tech already configured (provided in requirements)

Deployment

1. Clone and navigate to project:

   cd ai-order-parser

2. Deploy infrastructure:

   ./deploy.sh

3. Wait for DNS propagation (up to 24 hours)

4. Test the system: Send an email with PDF attachment to invoice-bot@katechat.tech

Local Testing

1. Configure environment variables:

cd processor
cp .env.example .env
# Edit .env with your AWS credentials

2. Test the parser with sample invoices:

# Test a specific PDF
python test_local.py test ../data/Rechnung-38110.pdf

# Test all sample PDFs
python test_local.py test-all

# Show PDF info (without AI parsing)
python test_local.py info ../data/42507956.pdf

📧 How It Works

1. Email Receipt: Send invoice PDF to invoice-bot@katechat.tech
2. SES Processing: AWS SES receives email and stores in S3
3. Lambda Trigger: SES triggers Lambda function automatically
4. PDF Extraction: Function extracts PDF attachments from email
5. AI Parsing: AWS Bedrock Nova model receives PDF directly for multimodal analysis
6. Response: Parsed JSON and original PDF sent back via email

📄 Output Format

The system extracts invoices into this JSON structure:

{
    "invoice_number": "INV-2024-001",
    "receiver_name": "John Doe",
    "receiver_address": "123 Main St, City, 12345 Country",
    "issuer_name": "ACME Corporation",
    "issuer_address": "456 Business Ave, Town, 67890 Country",
    "total": 250.75,
    "items": [
        {
            "title": "Product A",
            "quantity": "2",
            "price": 125.00
        },
        {
            "title": "Service Fee", 
            "quantity": "1",
            "price": 0.75
        }
    ]
}

🌍 Multi-language Support

The system understands invoices in:

• English: Invoice, Total, Items, etc.
• German: Rechnung, Rechnungsnummer, Gesamt, Summe, etc.

🏗️ Architecture

┌─────────────────┐    ┌──────────────┐    ┌─────────────────┐
│   Email Client  │───▶│  AWS SES     │───▶│  AWS Lambda     │
└─────────────────┘    └──────────────┘    └─────────────────┘
                              │                       │
                              ▼                       ▼
                       ┌──────────────┐    ┌─────────────────┐
                       │   AWS S3     │    │ AWS Bedrock     │
                       │  (Storage)   │    │  (Nova AI)      │
                       └──────────────┘    └─────────────────┘

🔐 Security Features

• IAM Policies: Minimal permissions for each service
• Email Authentication: SPF, DKIM, and DMARC configured
• Encryption: S3 bucket encrypted at rest
• Access Control: Only authorized email addresses can send via SES
• Temporary Storage: Emails deleted after 30 days

💰 Cost Estimation

Approximate AWS costs for moderate usage (100 invoices/month):

• Lambda: ~$0.10/month (100 invocations × 30s avg)
• SES: ~$1.00/month (200 emails × $0.10/1000)
• S3: ~$0.01/month (minimal storage, 30-day lifecycle)
• Bedrock Nova: ~$5.00/month (varies by token usage)
• Route53: ~$0.50/month (DNS queries)

Total: ~$6.61/month

🐛 Troubleshooting

Common Issues

1. Emails not received:

   • Check DNS propagation: dig MX katechat.tech
   • Verify SES domain verification in AWS console
   • Check spam folder

2. Lambda function errors:

   • Check CloudWatch logs: /aws/lambda/invoice-parser
   • Verify Bedrock model permissions
   • Ensure S3 bucket permissions

3. PDF parsing issues:

   • Test locally with test_local.py
   • Check if PDF is text-based (not scanned image)
   • Verify Bedrock Nova model availability in region

Monitoring

• CloudWatch Logs: Function execution details
• SES Metrics: Email delivery statistics
• Lambda Metrics: Invocation count, duration, errors

🔧 Configuration

Environment Variables (Lambda)

• REGION: Deployment region (default: eu-central-1)
• S3_BUCKET: S3 bucket for email storage (auto-generated)
• FROM_EMAIL: Sender email address (invoice-bot@katechat.tech)

Terraform Variables

• aws_region: AWS region for deployment
• domain_name: Domain for SES (katechat.tech)
• route53_zone_id: Route53 zone ID (provided)

📈 Scaling

The system is designed to scale automatically:

• Lambda: Scales to handle concurrent email processing
• SES: No limits on email volume (within SES limits)
• S3: Unlimited storage capacity
• Bedrock: Handles multiple simultaneous requests

🤝 Contributing

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open Pull Request

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙋‍♂️ Support

For issues and questions:

1. Check the troubleshooting section above
2. Review CloudWatch logs for errors
3. Test locally using the provided scripts
4. Open an issue in the GitHub repository

---

Built with ❤️ using AWS Lambda, Bedrock Nova, SES, and Terraform