🦊 PDF to Structured Data

Extract structured data from PDF documents using Google Gemini AI.

Overview

This API service leverages Google's Gemini multimodal AI to extract structured data from PDF documents without requiring predefined schemas. Gemini can automatically analyze both text and visual elements in PDFs and extract them into structured JSON format.

Key features:

• Simple PDF upload and processing
• Dynamic data extraction without predefined schemas
• Uses Gemini's visual understanding capabilities
• Modern, user-friendly frontend interface

Architecture

The application follows a simple design:

pdf-to-structured-data/
├── app/
│   ├── api/
│   │   ├── routes/
│   │   │   └── pdf_routes.py    # API endpoints for PDF processing
│   ├── services/
│   │   └── gemini_service.py    # Gemini API integration
│   ├── utils/
│   │   └── file_utils.py        # File handling utilities
│   ├── config.py                # Configuration settings
│   └── main.py                  # FastAPI application entry point
├── frontend/
│   ├── index.html               # Frontend UI
│   ├── scripts.js               # Frontend JavaScript
│   └── README.md                # Frontend documentation
├── temp_uploads/                # Temporary storage for PDFs
├── requirements.txt
└── README.md

How It Works

1. Upload: PDF documents are uploaded through the API or frontend
2. Process: PDF is sent to Gemini with a carefully crafted prompt
3. Extract: Gemini analyzes the document and extracts structured data
4. Return: Structured JSON data is returned to the client

Installation

Prerequisites

• Python 3.9+
• Google Gemini API key

Setup

1. Clone the repository:

git clone https://github.com/SlyyCooper/pdf-to-structured-data.git
cd pdf-to-structured-data

2. Install dependencies:

pip install -r requirements.txt

3. Create a .env file with your Gemini API key:

GOOGLE_API_KEY=your_api_key_here

Usage

Starting the API

uvicorn app.main:app --reload

The API will be available at http://localhost:8000

Running the Frontend

Open frontend/index.html in a web browser, or serve it using a static file server.

For example, using Python's built-in HTTP server:

cd frontend
python -m http.server 8080

Then open http://localhost:8080 in your browser.

API Endpoints

Upload a PDF

POST /api/v1/upload

• Request: Multipart form with file
• Response: JSON with file ID

Extract Data

POST /api/v1/extract/{file_id}

• Path Parameters:
  • file_id: ID of the uploaded file
• Request Body (optional):
  • document_type: Hint about the document type (optional)
• Response: JSON with extracted structured data

Health Check

GET /api/v1/health

• Response: API health status

Examples

1. Upload a PDF

curl -X POST -F "file=@invoice.pdf" http://localhost:8000/api/v1/upload

Response:

{
  "file_id": "3fa85f64-5717-4562-b3fc-2c963f66afa6.pdf",
  "filename": "invoice.pdf",
  "size_bytes": 12345
}

2. Extract Data

curl -X POST -H "Content-Type: application/json" -d '{"document_type": "invoice"}' http://localhost:8000/api/v1/extract/3fa85f64-5717-4562-b3fc-2c963f66afa6.pdf

Response:

{
  "document_type": "invoice",
  "vendor": {
    "name": "ACME Corp",
    "address": "123 Main St, Anytown, USA"
  },
  "invoice_number": "INV-12345",
  "date": "2023-06-01",
  "items": [
    {
      "description": "Product A",
      "quantity": 2,
      "unit_price": 49.99,
      "amount": 99.98
    }
  ],
  "subtotal": 99.98,
  "tax": 8.00,
  "total": 107.98
}

Frontend

The application includes a modern, user-friendly frontend designed with an Apple-inspired aesthetic. Key features:

• Drag & drop PDF uploads
• Clear visual feedback during processing
• Formatted JSON display with syntax highlighting
• One-click copy to clipboard
• Responsive design for mobile and desktop

For more details, see the Frontend README.

License

MIT License