AI导哈

AI辅导平台

一个离线优先的AI辅导系统，通过模型上下文协议（MCP）为难民和资源匮乏的学生提供个性化学习支持，支持本地网络访问、双语翻译与云端同步。

  • by Moniaar
  • 本地部署
  • 更新于
在 GitHub 查看 Moniaar/Ufuq

使用说明

Ufuq AI Tutoring Platform

An offline-first AI tutoring system using Model Context Protocol (MCP) for refugee and underserved students. Works locally via Wi-Fi hotspot with optional cloud integration.

🌟 Features

Core Capabilities

  • Offline-First: Works without internet using local AI models
  • Multi-Agent System: Specialized agents for tutoring, translation, quizzes, and content retrieval
  • Bilingual Support: Arabic and English with real-time translation
  • Progressive Web App: Works on phones, tablets, and computers
  • Local Network Access: Students connect via Wi-Fi hotspot or LAN
  • Cloud Sync: Syncs progress when internet is available

MCP Agents

  1. Tutor Agent - Explains concepts, answers questions
  2. Translator Agent - Arabic ↔ English translation
  3. Quiz Agent - Generates and grades assessments
  4. Content Agent - Retrieves from stored educational materials (RAG)
  5. Sync Agent - Syncs data to cloud when online

🚀 Quick Start

Prerequisites

  • Python 3.9+
  • 4GB+ RAM (for local models)
  • Wi-Fi hotspot capability or local network

Installation

# Clone the repository
git clone <your-repo-url>
cd ufuq

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Download models (first time only)
python download_models.py

Configuration

Create a .env file:

# Optional: Add API keys for online mode
OPENAI_API_KEY=your_key_here
SAMBANOVA_API_KEY=your_key_here
MODAL_API_KEY=your_key_here

# Debug mode
DEBUG=True

Running the Server

# Start the Flask backend
python app.py

The server will run on http://0.0.0.0:5000 (accessible on local network)

Connecting Students

  1. Enable Wi-Fi Hotspot on the host device
  2. Students connect to the hotspot
  3. Students open browser and navigate to: http://<host-ip>:5000
  4. Start learning!

📁 Project Structure

ufuq/
├── app.py                 # Main Flask application
├── mcp_runtime.py         # MCP agent orchestration hub
├── agents.py              # All 5 MCP agents
├── storage.py             # Database and caching
├── config.py              # Configuration management
├── requirements.txt       # Python dependencies
├── download_models.py     # Model download script
├── data/
│   ├── ufuq.db           # SQLite database
│   ├── cache/            # Offline cache
│   ├── vectordb/         # ChromaDB vector storage
│   └── content/          # Educational PDFs/materials
└── frontend/             # React PWA (separate)
    └── ...

🔧 API Endpoints

Chat (Tutor Agent)

POST /api/chat
{
  "student_id": "student_123",
  "message": "Explain Newton's laws",
  "language": "en"
}

Translation

POST /api/translate
{
  "text": "Hello, how are you?",
  "source_lang": "en",
  "target_lang": "ar"
}

Generate Quiz

POST /api/quiz/generate
{
  "student_id": "student_123",
  "topic": "Physics - Newton's Laws",
  "difficulty": "medium",
  "num_questions": 5,
  "language": "en"
}

Submit Quiz

POST /api/quiz/submit
{
  "student_id": "student_123",
  "quiz_id": 1,
  "answers": ["A", "B", "C", "D", "A"]
}

Retrieve Content (RAG)

POST /api/content/retrieve
{
  "query": "What is photosynthesis?",
  "subject": "biology",
  "language": "en"
}

Student Progress

GET /api/student/progress?student_id=student_123

Sync Data

POST /api/sync
{
  "student_id": "student_123"
}

Check Mode

GET /api/mode

🤖 MCP Architecture

How It Works

  1. Student Request → Flask receives HTTP request
  2. MCP Router → Routes to appropriate agent
  3. Agent Execution → Agent processes with local/cloud models
  4. Mode Detection → Automatically switches between offline/online
  5. Response → Returns result to student
  6. Storage → Saves interaction locally, syncs when online

Agent Selection Logic

# MCP Runtime automatically routes based on request type
mcp_runtime.route_request(
    agent_type="tutor",  # or "translator", "quiz", "content", "sync"
    payload={...}
)

Offline ↔ Online Switching

The system automatically detects internet connectivity:

  • Offline: Uses Hugging Face models (FLAN-T5, Helsinki-NLP)
  • Online: Uses OpenAI GPT-4 for better quality
  • Hybrid: Falls back to offline if online fails

📚 Adding Educational Content

Ingest PDFs for RAG

# Create a content ingestion script
from content_ingestion import ingest_pdf

# Add textbooks
ingest_pdf('path/to/physics_textbook.pdf', subject='physics')
ingest_pdf('path/to/math_textbook.pdf', subject='mathematics')

Manual Content Addition

from agents import ContentAgent
from config import Config

config = Config()
content_agent = ContentAgent(config)

# Add content manually
content_agent.collection.add(
    documents=["Newton's First Law: Objects at rest stay at rest..."],
    metadatas=[{"subject": "physics", "source": "textbook_ch1"}],
    ids=["physics_newton_1"]
)

🌐 Cloud Integration (Optional)

OpenAI Setup

# Set API key in .env
OPENAI_API_KEY=sk-...

Modal Setup (for serverless functions)

pip install modal
modal token new

SambaNova Setup

# Add SambaNova API key
SAMBANOVA_API_KEY=your_key

🧪 Testing

# Run tests
pytest tests/

# Test specific agent
pytest tests/test_tutor_agent.py

# Test offline mode
FORCE_OFFLINE=True pytest tests/

📱 Frontend Integration

The React PWA connects to these endpoints. Example:

// Chat with tutor
const response = await fetch('http://<host-ip>:5000/api/chat', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    student_id: 'student_123',
    message: 'Explain gravity',
    language: 'en'
  })
});

const data = await response.json();
console.log(data.response);  // AI tutor's explanation

🔐 Security Considerations

  • Local Network Only: By default, only accessible on LAN
  • No Authentication: Add authentication for production
  • Data Privacy: All data stored locally, synced optionally
  • API Keys: Keep in .env, never commit to git

🐛 Troubleshooting

Models Not Loading

# Re-download models
python download_models.py --force

Port Already in Use

# Change port in app.py
app.run(host='0.0.0.0', port=5001)

Out of Memory

  • Reduce model sizes in config.py
  • Use smaller models (e.g., flan-t5-small instead of base)
  • Disable local model loading and use online-only mode

Connection Issues

  • Check firewall settings
  • Ensure all devices on same network
  • Use IP address instead of localhost

📊 Performance Optimization

For Low-Resource Devices

# In config.py, use smaller models
'offline': {
    'tutor': 'google/flan-t5-small',  # Instead of base
    'translator': 'Helsinki-NLP/opus-mt-tc-big-en-ar'  # Smaller variant
}

Batch Processing

# Process multiple quiz questions at once
quiz_agent.execute({
    'action': 'batch_generate',
    'topics': ['physics', 'math', 'chemistry']
})

🤝 Contributing

  1. Fork the repository
  2. Create a feature branch
  3. Make your changes
  4. Test offline and online modes
  5. Submit a pull request

📄 License

MIT License - Feel free to use for educational purposes

🆘 Support

For issues during the hackathon:

  • Check logs: tail -f logs/ufuq.log
  • Test mode: curl http://localhost:5000/health

🚀 Next Steps

  • [ ] Add voice input/output
  • [ ] Implement peer-to-peer quiz sharing
  • [ ] Add gamification (points, badges)
  • [ ] Multi-user collaboration features
  • [ ] Teacher dashboard for monitoring
  • [ ] More languages (French, Spanish)
  • [ ] Offline video lessons

Built with ❤️ for underserved students worldwide

相关 MCP