🔍 Gnosis OCR

Enterprise-Grade GPU-Accelerated OCR Service

Extract text from documents with state-of-the-art AI models

Quick Start • Features • Documentation • Deployment • API

✨ Features

• 🚀 GPU-Accelerated Processing - Powered by NVIDIA CUDA for blazing-fast OCR
• 🎯 State-of-the-Art Models - Uses Nanonets OCR-s for superior accuracy
• 📄 Large File Support - Process up to 500MB PDFs with HTTP/2 streaming (uploads work during model loading)
• ⚡ Live Results Preview - See extracted text in real-time as pages complete
• 🔄 Incremental Processing - No waiting - results appear as they're ready
• ☁️ Cloud Native - Deploy to Google Cloud Run with auto-scaling
• 🔒 Enterprise Security - User isolation, session management, and audit trails
• 📊 Real-Time Progress - Live processing updates and detailed analytics
• 🌐 Modern Web UI - Beautiful, responsive interface with dark mode
• 🔌 REST API - Full programmatic access with OpenAPI docs

🚀 Quick Start

Prerequisites

• NVIDIA GPU with 12GB+ VRAM (RTX 3060 12GB/3070/3080/3090 or better)

• Docker Desktop with GPU support

• 16GB+ RAM (32GB recommended for large documents)

Local Development

# Clone the repository
git clone https://github.com/kordless/gnosis-ocr.git
cd gnosis-ocr

# Create environment files
# Copy the sample file for local development
cp .env.sample .env
# Use the same file for Cloud Run and edit PROJECT_ID if deploying
cp .env.sample .env.cloudrun

# Build and run with GPU support
docker-compose up --build

🎉 That's it! Visit http://localhost:7799 to start processing documents.

Cloud Deployment

Deploy to Google Cloud Run with NVIDIA L4 GPU:

# One-command deployment
./scripts/deploy.ps1 -Target cloudrun

See Deployment Guide for detailed instructions.

📸 Screenshots

Document Upload

Live Results Preview

OCR Results

🏗️ Architecture

graph TD
    A[User Upload] --> B[FastAPI Backend]
    B --> C[Document Processor]
    C --> D[GPU OCR Engine]
    D --> E[Text Extraction]
    E --> F[GCS Storage]
    F --> G[Results API]
    G --> H[Web UI]
    
    subgraph "GPU Processing"
        D --> I[Nanonets OCR-s Model]
        I --> J[CUDA Acceleration]
    end
    
    subgraph "Cloud Storage"
        F --> K[Model Cache]
        F --> L[User Data]
        F --> M[Session Files]
    end

Loading

🎯 Performance

Metric	Local (RTX 3090)	Cloud Run (L4)
Model Startup	~4 minutes	~30 seconds
Processing Speed	~20 sec/page	~5-10 sec/page
Live Preview	✅ Real-time	✅ Real-time
Max File Size	500MB	500MB
Concurrent Users	1-2	10+
Model Loading	Cached	Persistent Mount

🔧 Configuration

Environment Variables

Variable	Description	Default
DEVICE	Processing device	cuda
MODEL_NAME	OCR model to use	nanonets/Nanonets-OCR-s
MAX_FILE_SIZE	Maximum upload size	500MB
SESSION_TIMEOUT	Session duration	3600 seconds
LOG_LEVEL	Logging verbosity	INFO

GPU Requirements

• Minimum: 12GB VRAM (RTX 3060 12GB, RTX 3070)
• Recommended: 12GB+ VRAM (RTX 3080/3090/4070 Ti/4080/4090)
• Cloud: NVIDIA L4 (24GB VRAM)

📖 Documentation

• 📚 Deployment Guide - Complete deployment instructions
• 🔧 Setup Guide - Development environment setup
• 🌐 API Reference - Interactive OpenAPI documentation
• 🐳 Docker Guide - Container configuration and troubleshooting

🚀 Deployment

Local Development

docker-compose up --build

Local Dev Environment

./scripts/deploy.ps1 -Target local

Production (Cloud Run)

./scripts/deploy.ps1 -Target cloudrun

Cloud Build (CI/CD)

gcloud builds submit --config cloudbuild.yaml

📡 API Reference

Upload Document

curl -X POST "http://localhost:7799/api/v1/jobs/submit" \
  -H "Content-Type: multipart/form-data" \
  -F "file=@document.pdf"

Check Status (with Live Results)

curl "http://localhost:7799/api/v1/jobs/{job_id}/status"
# Returns incremental results in partial_results array

Get Final Results

curl "http://localhost:7799/api/v1/jobs/{job_id}/result"

Example Status Response

{
  "status": "processing",
  "progress": {
    "current_page": 3,
    "total_pages": 10,
    "percent": 30
  },
  "partial_results": [
    {
      "page_number": 1,
      "text": "# Document Title\nExtracted content...",
      "status": "completed",
      "confidence": 0.95
    },
    {
      "page_number": 2,
      "text": "Page 2 content...",
      "status": "completed",
      "confidence": 0.93
    }
  ]
}

Complete API documentation available at /docs endpoint.

🏢 Enterprise Features

• Live Results Streaming - Real-time text extraction as pages complete
• Incremental Processing - No waiting for entire documents to finish
• Multi-Tenant Architecture - User isolation and data partitioning
• Audit Logging - Complete request/response tracking
• Auto-Scaling - Handles traffic spikes automatically
• Health Monitoring - Built-in health checks and metrics
• Security - CORS, rate limiting, and input validation
• Storage Integration - Google Cloud Storage with lifecycle management

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

1. Fork the repository
2. Create a 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 a Pull Request

📄 License

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

🙏 Acknowledgments

• Nanonets - For the excellent OCR model
• HuggingFace - For the transformers library
• FastAPI - For the robust web framework
• NVIDIA - For CUDA and GPU acceleration

📊 Analytics

---

⭐ Star this repository if you find it useful! ⭐

Report Bug • Request Feature • Join Discord