AI Document Processor

A FastAPI service that extracts structured data from documents using OCR and LLM-based parsing. Upload a PDF or image, get clean JSON back with the relevant fields extracted.

What It Does

1. Accepts PDF, PNG, JPG, TIFF uploads via REST API or web UI
2. Extracts text using Tesseract OCR (with multi-language support) or PDF text layers
3. Classifies the document type automatically (invoice, receipt, contract, or generic)
4. Parses the text into structured JSON using any OpenAI-compatible LLM API
5. Returns clean, typed fields — amounts as numbers, dates in ISO format, null for missing data

Supported Document Types

Type	Extracted Fields
Invoice	Invoice number, dates, vendor/bill-to info, currency, line items with quantities and prices, subtotal/tax/total
Receipt	Store name/address, date, itemized list, subtotal, tax, total, payment method
Contract	Title, parties and roles, effective/expiration dates, key terms, governing law, signatures
Generic	Title, date, summary, key entities, key data points

Setup

Prerequisites

• Python 3.11+
• Tesseract OCR installed and on PATH
• An API key for any OpenAI-compatible LLM service

Install

cd ai-document-processor
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Configure

Create a .env file:

LLM_API_KEY=sk-your-api-key
LLM_API_BASE=https://api.openai.com/v1    # or any compatible endpoint
LLM_MODEL=gpt-4o-mini                      # or claude-3-haiku, etc.
OCR_LANG=eng                               # tesseract language codes
MAX_FILE_SIZE_MB=20

Run

python app.py
# or
uvicorn app:app --host 0.0.0.0 --port 8080 --reload

Open http://localhost:8080 for the web UI.

API Reference

POST /api/process

Process a single document.

Parameters:

Name	Type	Description
file	form-data	The document file (required)
lang	query	Tesseract language codes, e.g. eng+heb (default: eng)
doc_type	query	Force document type: invoice, receipt, contract, generic (default: auto-detect)

Response:

{
  "filename": "invoice-042.pdf",
  "extracted_text": "INVOICE #042...",
  "result": {
    "document_type": "invoice",
    "fields": {
      "invoice_number": "042",
      "date": "2026-03-01",
      "vendor": { "name": "Acme Corp", "address": "123 Main St" },
      "total": 1500.00,
      "line_items": [
        { "description": "Consulting", "quantity": 10, "unit_price": 150.00, "amount": 1500.00 }
      ]
    }
  }
}

POST /api/batch

Process up to 10 documents concurrently. Same query parameters as /api/process, but send multiple files under the files field.

Response:

{
  "count": 3,
  "results": [ ... ]
}

GET /health

Returns service status and whether the LLM key is configured.

Web UI

The built-in UI at / supports:

• Drag-and-drop or click-to-browse file selection
• Multiple file upload
• Language and document type selection
• Collapsible result cards showing extracted fields and raw OCR text

Architecture

Upload → File validation → OCR (Tesseract / PDF text layer)
       → LLM classification → LLM field extraction → JSON response

• OCR runs in a thread pool to avoid blocking the async event loop
• Batch processing uses asyncio.gather for concurrent file handling
• LLM calls go through httpx.AsyncClient to any OpenAI-compatible endpoint
• PDF handling tries the embedded text layer first, falls back to page-by-page OCR at 300 DPI

Project Structure

app.py            — FastAPI routes, OCR pipeline, file handling
processors.py     — Document classification, extraction prompts, LLM client
templates/
  index.html      — Web UI with drag-and-drop upload
requirements.txt  — Python dependencies
.env              — API keys and config (not committed)

License

MIT