Power up LLMs to be Factual Ninjas using GraphRAG
summary
GraphRAG claims to cut LLM hallucinations by combining vector embeddings stored as node properties with an extracted entity/knowledge graph so retrieval does multi‑hop reasoning and returns traceable sources. Implementation includes a no‑code LLM graph builder, a Python package for graph‑RAG pipelines, and an Aura cloud deployment; demos use a 12,000‑movie subset and benchmarks report better accuracy than vector‑only RAG. It supports precomputed embeddings, fixed‑size chunking, fine‑grained access control and a free Neo4j Aura tier of 200,000 nodes/400,000 relationships.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:10GraphRAG overview
- 2:35Graph databases and theory
- 5:23Neo4j graph platform
- 8:55Limitations of vector-only RAG
- 13:31GraphRAG: vectors + graphs
- 15:59Building knowledge graphs
- 18:25GraphRAG with RFP example
- 23:23GraphRAG data sources
- 24:47LLM Graph Builder demo
- 28:41GraphRAG Python package
- 31:10Neo4j Aura Agent
- 32:12Key resources
- 36:25Q&A
- 39:37Q&A
- 48:58Closing
related talks
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
28mEnhancing LLMs with Graph Technology
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
36mAgentic GraphRAG: AI’s Logical Edge
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs
52mGive Your LLMs a Left Brain
36mGraphRAG Explained 🔥 Building Smarter LLMs
32mGraphRAG: Powering Up LLMs with Knowledge Graphs