Self-RAG Pipeline with LangGraph

$ cat ./self-rag-pipeline/README.md

// self-evaluating rag pipeline — the llm grades its own retrieved docs, checks for hallucinations, and validates its answer before returning it. built as a stateful langgraph workflow.

// stack: langgraph · langchain · groq · chroma · huggingface · python

// graph

START
  ↓
init_groq_model          // load llama 3.3 70b via groq
  ↓
prepare_vector_database  // scrape urls → chunk → embed → chroma
  ↓
fetch_relevant_docs      // vector similarity search
  ↓
filter_docs_by_relevance // llm grades each doc: yes or no
  ↓ (conditional)
  ├── no relevant docs → END
  └── docs found ↓
        produce_answer       // rag prompt → answer draft
          ↓
        detect_hallucination // is answer grounded in facts?
          ↓
        validate_answer      // does answer address the question?
          ↓
        END