Agentic GraphRAG: AI’s Logical Edge
Talkat Open Source Summit Japan + AI_dev: Open Source GenAI & ML Summit + Automotive Linux Summit 2025 · 2025-12-08..12-10·
summary
Claims that combining vector search with a knowledge graph reduces hallucination and improves multi‑hop reasoning, demonstrated with a 36‑student grid puzzle where LLMs give wrong maxima versus human solution, and a healthcare case where vector RAG yields generic care but graph‑enhanced RAG returns medication plus pulmonary rehab. Demo uses Neo4j, a Knowledge Graph Builder UI, and an MCP toolchain to run GraphRAG retrieval patterns and MCP tools that execute Cypher and enforce role‑based access.
AI-written summary · about
chapters (10)
- 0:00AI project failures
- 1:07Logic problem on class seating
- 6:51LLM reasoning limits
- 10:22Solving with GraphRAG
- 16:43GraphRAG demo with knowledge graphs
- 20:14AI agents and benefits
- 23:02Agentic system architecture
- 26:43MCP servers as tools
- 30:16MCP cipher tool demo
- 33:30Summary and resources
related talks
30mAgentic RAG with Knowledge Graphs Explained
27mWhen Vectors Aren’t Enough: Unlocking AI’s Hidden Connections with Graphs
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
23mAgentic GraphRAG with MAGMA Manifold Agency with Graph Models and Augmentation
27mLies, Damn Lies, and AIs
49mPower up LLMs to be Factual Ninjas using GraphRAG
20mBeyond RAG: Building the AI Corporate Brain with Knowledge Graphs and Virtue Ethics
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs