GraphRAG: Powering Up LLMs with Knowledge Graphs
summary
GraphRAG shows how Neo4j combines knowledge graphs with RAG: it ingests SEC EDGAR Form 10‑K (300–400 page PDFs) and Form 13 filings, extracts entities and embeddings, and links structured records to text chunks so queries that vector-only RAG fails on become answerable. A demo on EDGAR and Melbourne metro pages generates 358 nodes and 304 relationships, and a vector+graph retriever returns named asset managers where pure vector search does not. The project includes an open‑source llm graph builder and a graph retriever + vector search stack that runs on Neo4j (Aura free: 200k nodes / 400k relationships).
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:13What is a graph
- 8:53Neo4j capabilities
- 10:59Graph use cases
- 11:20Limitations of vector-only RAG
- 14:55Introducing GraphRAG
- 19:51GraphRAG architecture
- 20:12Demo: vector vs GraphRAG
- 24:53Demo: LLM Graph Builder
- 28:42Governance in GraphRAG
- 29:24GraphRAG reference architecture
- 30:30Resources
related talks
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
49mPower up LLMs to be Factual Ninjas using GraphRAG
36mGraphRAG Explained 🔥 Building Smarter LLMs
30mAgentic RAG with Knowledge Graphs Explained
36mGraphRAG for Law: Building Legal Reasoning Agents