README.md

metadata

title: Corrective RAG (LangGraph)
emoji: 🚀
colorFrom: blue
colorTo: gray
sdk: streamlit
sdk_version: 1.32.0
app_file: app.py
pinned: false

Corrective RAG (CRAG) with LangGraph

This project implements a Corrective Retrieval-Augmented Generation (CRAG) system using LangGraph. It is designed to be a self-correcting RAG agent that actively evaluates the quality of retrieved documents and generated answers, actively mitigating hallucinations.

🚀 Key Features

• Self-Correction: The agent evaluates retrieved documents for relevance. If documents are irrelevant, it automatically falls back to a web search.
• Active Routing: Routes questions intelligently between vector store retrieval and web search based on the query intent.
• Hallucination Grading: Checks if the generated answer is grounded in the documents and if it actually answers the user's question.
• Multilingual Support: The Streamlit interface supports both Turkish and English.

🛠️ Tech Stack

• Orchestration: LangGraph
• LLM: DeepSeek (via DeepSeek-V3 / deepseek-chat)
• Embeddings: Hugging Face (sentence-transformers/all-MiniLM-L6-v2)
• Vector Database: ChromaDB
• Web Search: Tavily Search API
• UI: Streamlit

🌐 Live Demo

Verify the application directly on the web: 👉 Click here to use the application (If the agent doesn't respond, it could mean that the quota for the LLM being used has been exhausted 🙂)

⚙️ Setup & Installation

1. Clone the repository:

   git clone https://github.com/MikailBitik/CorrectiveRAGProject.git
   cd corrective-rag-project
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Environment Variables: Create a .env file in the root directory and add your API keys:

   DEEPSEEK_API_KEY=your_deepseek_key
   HF_TOKEN=your_huggingface_token #for using embedding models from huggingface
   TAVILY_API_KEY=your_tavily_key
   LANGCHAIN_API_KEY=your_langchain_key # Optional: for LangSmith tracing
   LANGCHAIN_TRACING_V2=true           # Optional
   LANGCHAIN_PROJECT=CorrectiveRAGProject # Optional
   

🏃 Usage

1. Ingest Data: First, crawl and index the knowledge base (default: Lilian Weng's blog posts).

python ingestion.py

2. Run the Application: Start the Streamlit interface.

streamlit run app.py