AI-powered document extraction running 100% locally
DocExtract is a full-stack web application that allows users to upload business documents (invoices, contracts, receipts), automatically extract structured information using OCR and local LLM, review/edit extracted fields, and export data as CSV or JSON.
🔒 Privacy First: All processing happens locally. No cloud APIs, no data leaves your infrastructure.
┌─────────────────────────────────────────────────────────────────────────────┐
│ DocExtract System │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────────┐ ┌─────────────────────────────────────────┐ │
│ │ │ │ Backend (FastAPI) │ │
│ │ Frontend │ HTTP │ ┌─────────┐ ┌─────────┐ ┌─────────┐ │ │
│ │ (Next.js) │◄───────►│ │ Upload │ │Documents│ │ Export │ │ │
│ │ │ API │ │ Router │ │ Router │ │ Router │ │ │
│ │ ┌───────────┐ │ │ └────┬────┘ └────┬────┘ └────┬────┘ │ │
│ │ │ Dashboard │ │ │ │ │ │ │ │
│ │ ├───────────┤ │ │ ▼ ▼ ▼ │ │
│ │ │ Upload │ │ │ ┌─────────────────────────────────┐ │ │
│ │ ├───────────┤ │ │ │ Services Layer │ │ │
│ │ │ Documents │ │ │ │ ┌─────────┐ ┌─────────────┐ │ │ │
│ │ ├───────────┤ │ │ │ │ OCR │ │ LLM │ │ │ │
│ │ │ Groups │ │ │ │ │ Service │ │ Service │ │ │ │
│ │ ├───────────┤ │ │ │ │(Tesser.)│ │(Ollama/HF) │ │ │ │
│ │ │ Export │ │ │ │ └────┬────┘ └──────┬──────┘ │ │ │
│ │ └───────────┘ │ │ │ │ │ │ │ │
│ │ │ │ │ ▼ ▼ │ │ │
│ │ Tailwind CSS │ │ │ ┌─────────────────────────┐ │ │ │
│ │ React │ │ │ │ Extraction Service │ │ │ │
│ └─────────────────┘ │ │ │ (Pipeline Orchestrator)│ │ │ │
│ │ │ │ └────────────┬────────────┘ │ │ │
│ │ │ └───────────────┼────────────────┘ │ │
│ │ │ │ │ │
│ │ │ ▼ │ │
│ │ │ ┌─────────────────────────────────┐ │ │
│ │ │ │ Database (SQLite/Postgres) │ │ │
│ │ │ │ ┌──────────┐ ┌─────────────┐ │ │ │
│ │ │ │ │Documents │ │ProcessingLog│ │ │ │
│ │ │ │ └──────────┘ └─────────────┘ │ │ │
│ │ │ └─────────────────────────────────┘ │ │
│ │ └────────────────────────────────────────┘ │
│ │ │
│ │ ┌────────────────────────────────────────────────────┐ │
│ └────►│ File Storage │ │
│ │ (./uploads/) │ │
│ └────────────────────────────────────────────────────┘ │
│ │
└────────────────────────────────────────────────────────────────────────────┘
Processing Pipeline:
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ Upload │───►│ OCR │───►│ Clean │───►│ LLM │───►│ Store │
│ File │ │ Extract │ │ Text │ │ Analyze │ │ Result │
└──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘
- 📤 Multi-file Upload: Drag & drop up to 20 documents at once
- 🔍 OCR Processing: Extract text from PDF, JPG, PNG using Tesseract
- 🤖 AI Analysis: Local LLM extracts structured fields (Ollama or HuggingFace)
- 📊 Document Types: Automatically classify as Invoice, Contract, Receipt
- ✏️ Review & Edit: View OCR text and edit extracted fields
- 📁 Smart Grouping: Group documents by type, vendor, or date
- 📥 Export: Download data as CSV or JSON
- 📈 Dashboard: Overview with statistics and confidence metrics
- 🔐 100% Local: No cloud APIs, all processing on your machine
- Python 3.10+
- Node.js 18+
- Tesseract OCR installed
- Poppler (for PDF processing)
- 8GB+ RAM recommended
- Option A: Ollama running locally with a model (e.g., llama3.2, mistral)
- Option B: HuggingFace Transformers (TinyLlama or Phi-2)
git clone https://github.com/yourusername/docextract.git
cd docextractWindows:
# Install Tesseract
winget install UB-Mannheim.TesseractOCR
# Install Poppler (for PDF support)
# Download from: https://github.com/oschwartz10612/poppler-windows/releases
# Add bin folder to PATH
# Install Ollama (for LLM)
winget install Ollama.OllamamacOS:
brew install tesseract poppler
brew install ollamaLinux (Ubuntu/Debian):
sudo apt update
sudo apt install tesseract-ocr poppler-utils
# Install Ollama: curl -fsSL https://ollama.com/install.sh | shcd backend
# Create virtual environment
python -m venv venv
# Activate virtual environment
# Windows:
venv\Scripts\activate
# macOS/Linux:
source venv/bin/activate
# Install dependencies
pip install -r requirements.txt
# Copy environment file
copy .env.example .env # Windows
cp .env.example .env # macOS/Linux
# Edit .env as needed (especially LLM settings)cd frontend
# Install dependencies
npm install# Pull a model (run Ollama first)
ollama pull llama3.2
# Or for smaller model:
ollama pull tinyllamacd backend
# Activate virtual environment first
# Windows:
venv\Scripts\activate
# Then run:
python main.py
# Or with uvicorn:
uvicorn main:app --reload --host 0.0.0.0 --port 8000Backend will be available at: http://localhost:8000 API Documentation: http://localhost:8000/docs
cd frontend
npm run devFrontend will be available at: http://localhost:3000
ollama servedocextract/
├── backend/
│ ├── main.py # FastAPI application entry point
│ ├── config.py # Configuration settings
│ ├── database.py # Database connection
│ ├── schemas.py # Pydantic schemas
│ ├── utils.py # Utility functions
│ ├── models/
│ │ ├── __init__.py
│ │ └── document.py # SQLAlchemy models
│ ├── routers/
│ │ ├── __init__.py
│ │ ├── upload.py # Upload endpoints
│ │ ├── documents.py # Document CRUD endpoints
│ │ ├── groups.py # Grouping endpoints
│ │ └── export.py # Export endpoints
│ ├── services/
│ │ ├── __init__.py
│ │ ├── ocr_service.py # Tesseract OCR
│ │ ├── llm_service.py # Local LLM integration
│ │ └── extraction_service.py # Pipeline orchestration
│ ├── requirements.txt
│ └── .env.example
├── frontend/
│ ├── src/
│ │ ├── app/
│ │ │ ├── layout.tsx
│ │ │ ├── page.tsx # Landing page
│ │ │ ├── globals.css
│ │ │ └── dashboard/
│ │ │ ├── layout.tsx
│ │ │ ├── page.tsx # Dashboard
│ │ │ ├── upload/
│ │ │ ├── documents/
│ │ │ ├── groups/
│ │ │ └── export/
│ │ ├── components/
│ │ │ ├── Sidebar.tsx
│ │ │ ├── FileUpload.tsx
│ │ │ ├── DocumentTable.tsx
│ │ │ └── StatsCards.tsx
│ │ └── lib/
│ │ ├── api.ts # API client
│ │ └── utils.ts # Utility functions
│ ├── package.json
│ ├── tailwind.config.js
│ └── next.config.js
├── samples/
│ ├── sample_invoice.txt
│ ├── sample_contract.txt
│ └── sample_receipt.txt
└── README.md
Edit backend/.env to configure:
# LLM Provider: "ollama" or "huggingface"
LLM_PROVIDER=ollama
# Ollama settings
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama3.2
# HuggingFace settings (if using)
HUGGINGFACE_MODEL=TinyLlama/TinyLlama-1.1B-Chat-v1.0
# Processing
CONFIDENCE_THRESHOLD=70.0| Method | Endpoint | Description |
|---|---|---|
| POST | /api/upload/ |
Upload multiple files |
| GET | /api/documents/ |
List documents (with filters) |
| GET | /api/documents/{id} |
Get document details |
| PUT | /api/documents/{id} |
Update document fields |
| POST | /api/documents/{id}/verify |
Mark as verified |
| POST | /api/documents/{id}/reprocess |
Reprocess document |
| DELETE | /api/documents/{id} |
Delete document |
| GET | /api/documents/stats |
Get dashboard statistics |
| GET | /api/groups/ |
Get grouped documents |
| GET | /api/export/json |
Export as JSON |
| GET | /api/export/csv |
Export as CSV |
The samples/ folder contains example documents:
sample_invoice.txt- Invoice from ACME Corporationsample_contract.txt- Service Agreement Contractsample_receipt.txt- Store receipt
Convert to PDF/image for testing, or use them to understand expected extraction.
- Ensure Tesseract is installed and in PATH
- On Windows, you may need to set
TESSERACT_CMDin.env
- Ensure Poppler is installed and in PATH
- On Windows, add Poppler's bin folder to PATH
- Ensure Ollama is running (
ollama serve) - Try a smaller model (
tinyllama) - Increase
LLM_TIMEOUTin.env
- Ensure document images are clear and high resolution
- Try increasing
PDF_DPIin config - Documents with poor formatting may need manual review
MIT License - See LICENSE file for details.