💬 LiveKit RAG Assistant v2.0
Enterprise-grade AI semantic search + real-time web integration for LiveKit documentation
🎯 Features
- Dual Search: Pinecone docs (3,000+ vectors) + Tavily real-time web
- Standard MCP: Async LangChain with Model Context Protocol
- Ultra-Fast: Groq LLM (llama-3.3-70b) sub-5s responses
- Premium UI: Glassmorphism design with 60+ animations
- Source Attribution: Full transparency on every answer
🚀 Quick Start
# Setup
conda create -n langmcp python=3.12
conda activate langmcp
pip install -r requirements.txt
# Configure .env
GROQ_API_KEY=your_key
TAVILY_API_KEY=your_key
PINECONE_API_KEY=your_key
PINECONE_INDEX_NAME=livekit-docs
# Terminal 1: Start MCP Server
python mcp_server_standard.py
# Terminal 2: Start UI
streamlit run app.py
App opens at http://localhost:8501
🏗️ Architecture
Streamlit (app.py) → MCP Server → Dual Search:
├─ Pinecone: Semantic search on embeddings (384-dim)
└─ Tavily: Real-time web results
↓
Groq LLM (2048 tokens, temp 0.3) → Response + Sources
🔧 Tech Stack
| Layer | Tech | Purpose |
|---|---|---|
| Frontend | Streamlit | Premium glassmorphism UI |
| Backend | MCP Standard | Async subprocess |
| LLM | Groq API | Ultra-fast inference |
| Embeddings | HuggingFace | all-MiniLM-L6-v2 (384-dim) |
| Vector DB | Pinecone | Serverless similarity search |
| Web Search | Tavily | Real-time internet results |
📚 Usage
- Choose mode: 📚 Docs or � Web
- Ask naturally: "How do I set up LiveKit?"
- Get instant answer with 📄 sources
- Copy messages or re-ask from history
⚡ Performance
- First query: ~15-20s (model load)
- Cached queries: 2-5s
- Search latency: <500ms
🛠️ Configuration
GROQ_API_KEY=gsk_***
TAVILY_API_KEY=tvly_***
PINECONE_API_KEY=***
PINECONE_INDEX_NAME=livekit-docs
🔄 Populate Docs
python ingest_docs_quick.py # Creates 3,000+ vector chunks
📊 Files
app.py- Streamlit UI with premium designmcp_server_standard.py- MCP server with toolsingest_docs_quick.py- Document ingestionrequirements.txt- Dependencies.env- API keys
🚨 Troubleshooting
| Issue | Solution |
|---|---|
| No results | Try web mode or different keywords |
| MCP not found | Start mcp_server_standard.py in Terminal 1 |
| Slow first response | Normal (15-20s) - model initializes once |
| API errors | Verify all keys in .env file |
� Features
✅ Real-time chat with 60+ animations ✅ Semantic + keyword hybrid search ✅ Copy-to-clipboard for messages ✅ Recent query suggestions ✅ System status dashboard ✅ Chat history persistence ✅ Query validation + error handling
Version: 2.0 | Status: ✅ Production Ready | Created: November 2025
👨💻 By @THENABILMAN | � Open Source | ❤️ For Developers