A full-stack Retrieval-Augmented Generation (RAG) agent combining a FastAPI backend (Python) with a modern React + TypeScript frontend. The backend leverages FAISS (and optionally Pinecone) for vector-based retrieval, integrates with LLMs (e.g., OpenAI GPT), and exposes APIs and websockets for conversational and retrieval tasks. The frontend provides a sleek, real-time chat interface.
The RAG Agent enables intelligent, context-aware conversations by augmenting large language models with efficient retrieval from a vector store. It is designed for applications such as recipe generation, knowledge assistants, or any scenario where grounding LLMs with domain data is valuable.
Backend (FastAPI, Python):
- Retrieval-augmented generation using FAISS or Pinecone vector stores
- Integration with OpenAI LLMs for response generation and embeddings
- Modular service and repository layers (e.g., for drinks, ingredients)
- WebSocket and REST API endpoints (
/retrieve,/generate) - Environment-based configuration, scalable and extensible design
Frontend (React, TypeScript, Vite):
- Real-time chat UI with WebSocket support
- Modern, responsive design using Ant Design components
- Type-safe, modular codebase
git clone <repo-url>
cd rag-agent- Python 3.8+
- uv (for Python package management)
- (Optional) Docker for containerized deployment
cd backend
# Create a virtual environment
uv venv .venv- Copy
.env.exampleto.envand fill in required values (API keys, etc.)
- By default, uses SQLite and FAISS. Pinecone can be enabled via environment variables.
uv run uvicorn --app-dir app main:app --reload- API: http://localhost:8000
- WebSocket: ws://localhost:8000
- Node.js 16+
- pnpm (recommended), npm, or yarn
cd ../frontend
pnpm install # or npm install or yarn installpnpm dev # or npm run dev or yarn devflowchart TB
UserInput[User Input] --> Prompt1[Prompt]
UserInput --> CheckIsFollowUp
subgraph CheckIsFollowUp[Check is follow up]
Prompt2[Prompt] --> LLM1((LLM)) --> Parser1[Parser]
end
CheckIsFollowUp --> FAISS["FAISS (Vector Store)"] -.-> Prompt1
OutputModel[Output Model] --> FormatInstruction[Format Instruction] --> Prompt1
MessageHistory[Message History] --> Prompt1
MessageHistory --> CheckIsFollowUp
Prompt1 --> LLM2((LLM)) --> Parser2[Parser] --> Result
- Backend: FastAPI, FAISS, Pinecone, SQLAlchemy, LangChain, OpenAI, uv
- Frontend: React, TypeScript, Vite, Ant Design, Socket.IO
MIT (or specify your license)