🤖 Agentic Invoice Processor

A beautiful, demo-ready invoice processing system built with Streamlit and LangGraph that uses AI agents to extract structured data from invoice images.

✨ Features

• 🤖 Pure Agentic Architecture: Built with LangGraph for intelligent, multi-step invoice processing
• 🔍 OCR Integration: Uses EasyOCR for text extraction from invoice images
• 🧠 AI-Powered Processing: GPT models clean OCR text and extract structured data
• ✅ Validation: Automatic validation of extracted invoice data
• 💾 SQLite Database: Persistent storage of processed invoices
• 📊 Analytics Dashboard: Visual insights into processed invoices
• 🎨 Beautiful UI: Modern Streamlit interface with real-time processing feedback

🚀 Quick Start

Prerequisites

• Python 3.8+
• OpenAI API key

Installation

1. Clone or navigate to the project directory:

cd invoice_generator

2. Create a virtual environment (recommended):

python -m venv invoice-env
source invoice-env/bin/activate  # On Windows: invoice-env\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Set up environment variables: Create a .env file in the project root:

OPENAI_API_KEY=your_openai_api_key_here

Or set it directly:

export OPENAI_API_KEY=your_openai_api_key_here  # On Windows: set OPENAI_API_KEY=your_key

5. Run the Streamlit app:

streamlit run app.py

The app will open in your browser at http://localhost:8501

📖 Usage

Processing Invoices

1. Upload an Invoice:

   • Go to the "Upload & Process" tab
   • Click "Choose an invoice image" and select a PNG/JPG file
   • Click "🚀 Process Invoice"

2. View Results:

   • See extracted data (vendor, invoice number, date, total, etc.)
   • Check processing steps status
   • View validation results
   • Inspect cleaned text and raw processing data

3. Browse Invoices:

   • Go to "View Invoices" tab
   • See all processed invoices in a table
   • Select an invoice to view detailed information
   • Delete invoices if needed

4. Analytics:

   • Go to "Analytics" tab
   • View statistics and charts
   • See trends and insights

🏗️ Architecture

Agentic Workflow

The system uses LangGraph to create an agentic pipeline:

OCR Text → Clean Text → Extract Data → Validate → Format Output

Each step is handled by a specialized agent node:

1. Clean OCR Node: Fixes OCR errors and normalizes text
2. Extract Data Node: Extracts structured JSON from cleaned text
3. Validate Node: Validates extracted data for completeness
4. Format Output Node: Prepares final output

Project Structure

invoice_generator/
├── app.py                 # Streamlit UI application
├── agents/
│   └── invoice_agent.py   # LangGraph agent implementation
├── utils/
│   ├── ocr.py            # OCR utilities (EasyOCR)
│   └── database.py       # SQLite database operations
├── requirements.txt       # Python dependencies
├── README.md             # This file
└── invoice.db            # SQLite database (created automatically)

🔧 Configuration

Model Selection

You can change the AI model in the Streamlit sidebar:

• gpt-4o-mini (default, fast and cost-effective)
• gpt-4o (more accurate, slower)
• gpt-3.5-turbo (fastest, less accurate)

Database

The SQLite database (invoice.db) is created automatically. It stores:

• Invoice metadata (vendor, number, date, total, currency)
• Extracted JSON data
• Validation status and errors
• Processing timestamps

🎯 Key Components

InvoiceProcessingAgent

The core agent class that orchestrates the LangGraph workflow:

from agents.invoice_agent import InvoiceProcessingAgent

agent = InvoiceProcessingAgent(model_name="gpt-4o-mini")
result = agent.process(file_name="invoice.png", ocr_text="...")

OCRReader

EasyOCR wrapper for text extraction:

from utils.ocr import OCRReader

reader = OCRReader()
text = reader.read_text("invoice.png")

InvoiceDB

Database operations for invoice storage:

from utils.database import InvoiceDB

db = InvoiceDB()
row_id = db.insert_invoice(file_name, data, raw_json)
invoices = db.get_all_invoices()

📊 Example Output

The agent extracts structured data like:

{
  "vendor": "Tech Solutions Ltd.",
  "number": "INV-1001",
  "date": "2025-09-10",
  "total": 553.88,
  "currency": "EUR",
  "line_items": [
    {
      "description": "Software License",
      "quantity": 1,
      "unit_price": 469.39,
      "amount": 469.39
    }
  ]
}

🛠️ Troubleshooting

OCR Issues

• Ensure images are clear and readable
• Supported formats: PNG, JPG, JPEG
• First run may take longer as EasyOCR downloads models

API Issues

• Verify your OpenAI API key is set correctly
• Check your API quota and billing
• Try a different model if one fails

Database Issues

• Delete invoice.db to reset the database
• Check file permissions in the project directory

🚧 Future Enhancements

• Batch processing of multiple invoices
• Export to CSV/Excel
• Multi-language support
• Custom validation rules
• Webhook notifications
• Cloud storage integration

📝 License

This project is open source and available for demonstration purposes.

🤝 Contributing

Feel free to fork, modify, and enhance this project!

---

Built with ❤️ using Streamlit, LangGraph, and OpenAI