AI KYC OCR ID Validator

AI-powered offline KYC document extraction and matching using Tesseract OCR and local LLMs via Ollama.

Features

• 📤 Upload scanned ID card images (JPEG/PNG)
• 🧠 Tesseract OCR with multi-language support (bul+eng)
• 🤖 Mistral 7B via Ollama for structured KYC field extraction
• ✅ JSON output with full name, DOB, nationality, document number
• 📊 Match percentage scoring against expected profiles
• 💻 Local processing — no cloud dependencies
• 🐳 Docker Compose support for frontend/backend services

Preview

Project Structure

ai-kyc-ocr-id-validator/
├── kyc-ocr-frontend/       # React + Redux + Tailwind CSS
├── kyc-ocr-backend/        # Node.js + Express + OCR + LLM
├── docs/                   # Architecture, prompts, benchmarks
├── docker-compose.yml
└── README.md

Prerequisites

• Node.js 18+
• Tesseract OCR
• Ollama with Mistral model
• Docker and Docker Compose

Quick Start

1. Pull the Mistral model:

ollama pull mistral

2. Start the application:

docker-compose up

3. Access the services:

• Frontend: http://localhost:5173
• Backend: http://localhost:5000

Development

• Source code changes are automatically reflected via volume mounting
• Rebuild containers after dependency changes:

docker-compose up --build

• Development mode with hot-reloading:

docker-compose -f docker-compose.dev.yml up

• Check logs:

docker-compose logs -f

System Architecture

graph TD
    subgraph Frontend
        A[React App] --> B[Image Upload]
        B --> C[Redux State]
        C --> D[Results Display]
    end

    subgraph Backend
        E[Express Server] --> F[Image Processing]
        F --> G[Tesseract OCR]
        G --> H[Text Extraction]
        H --> I[Ollama LLM]
        I --> J[Structured Data]
    end

    subgraph Local Services
        K[Tesseract OCR] --> L[Multi-language Support]
        M[Ollama] --> N[Mistral 7B]
    end

    B --> E
    J --> C
    G --> K
    I --> M

    style Frontend fill:#f9f,stroke:#333,stroke-width:2px
    style Backend fill:#bbf,stroke:#333,stroke-width:2px
    style Local Services fill:#bfb,stroke:#333,stroke-width:2px

Loading

Flow Description

1. Frontend Layer

   • User uploads ID image through React interface
   • Redux manages application state
   • Results displayed with match scoring

2. Backend Layer

   • Express server handles requests
   • Image processing and OCR extraction
   • LLM analysis for structured data

3. Local Services

   • Tesseract OCR for text extraction
   • Ollama running Mistral 7B for data structuring

Contributing

Please read CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.

Testing

# Compare different LLM prompts
npm run compare-prompts

# Run tests
npm test

Mock data available in mock/ directory:

• personal_image.png - Sample ID card