Graph RAG

In one focused session, you’ll add relationships to your RAG. We’ll auto-build a tiny knowledge graph (entities + links) from your docs and combine it with vector search to answer multi-hop, disambiguated questions with path-level citations.

No heavy infra required: we’ll use pure-Python first (NetworkX + FAISS) and show how to swap in Neo4j later.

Prerequisites: Basic Python, CLI (conda/pip), RAG familiarity, basic LLM's understanding.

Learning Outcomes

• Understand why Graph RAG: disambiguation, multi-hop, source clustering
• Build a mini knowledge graph from your docs (entities, co-occurrence edges)
• Run hybrid retrieval: vector search + graph neighborhood expansion
• Generate answers with path-aware citations (“A → B because edge X”)
• Know when to switch to Neo4j and where to put schema constraints