Agentic RAG with Knowledge Graphs Explained
summary
Argues that vector-only RAG fails at multi-hop reasoning and shows a production path where Neo4j stores vectors natively, exposes 70+ graph algorithms, and combines vector, keyword and graph traversal so retrieval links chunks and entities for traceable answers. Demonstrates two MCP servers — Neo4j Cipher (natural language → Cypher, e.g., heatmaps) and Neo4j Memory (persistent conversation facts that auto-link) — both integrated with Aura DB and agent SDKs; live demo had SSO hiccups but features are implemented.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:04Knowledge graphs and graph DBs
- 3:53Neo4j features and use cases
- 4:57Vector RAG limits and graph RAG intro
- 8:29Graph RAG vs vector RAG
- 10:57Building KGs from documents
- 13:48Access control in graph RAG
- 15:30Integration with generative AI models and frameworks
- 16:34AI agents and components
- 19:24Model Context Protocol benefits
- 21:31Agentic graph RAG with MCP
- 22:35MCP Neo4j Cipher server demo
- 24:43MCP Neo4j Memory server
- 27:10Agent SDKs and use cases
- 27:52Resources and next steps
- 28:55Closing remarks
related talks
27mWhen Vectors Aren’t Enough: Unlocking AI’s Hidden Connections with Graphs
39mEnhancing RAG with Neo4j Knowledge Graph
36mAgentic GraphRAG: AI’s Logical Edge
19mBeyond Retrieval: Adding a Reasoning Layer to RAG
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
28mData-Empowered AI Assistants
32mGraphRAG: Powering Up LLMs with Knowledge Graphs