topicexpert★ Pro

GraphRAG

Knowledge graph + RAG — know relationships between chunks, answer multi-hop questions.

4 hours2 resources1 prereqs

Classic RAG retrieves independent chunks — fails on multi-hop ("CEO of company X's family") questions.

GraphRAG (Microsoft Research): extract entities + relations from docs → build knowledge graph → find communities (Leiden) → summarize each community. Query = graph traversal + community summaries + chunk RAG.

Stack: Microsoft GraphRAG (Python), LangChain + Neo4j, LlamaIndex KnowledgeGraph.

Prerequisites

RAG Basics

Embedding + vector store + retrieve + augment — inject info the model doesn't know.

→

Resources(2)

GGitHub(1)

Microsoft GraphRAG

· en

freeofficial

PPaper(1)

From Local to Global — GraphRAG paper

· en

free

Re-ranking (Cross-Encoder)

Cost Monitoring

Open the full interactive roadmap