When Vectors Aren’t Enough: Unlocking AI’s Hidden Connections with Graphs
summary
Argues that vector-only RAG misses multi-hop, relational queries and shows a hybrid architecture: a knowledge graph (Neo4j) paired with a vector store and accessed by agents via MCP, plus agent memory using context graphs. Provides a GitHub repo and a Vercel demo; a live demo fails due to network but code is available. Cites research: data.world finds 3× higher response quality on 43 questions with GraphRAG, LinkedIn research reports a 28.6% decrease in issue resolution time, and Microsoft work on query-focused summarization.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:18Vector-only RAG limits
- 9:57What is a knowledge graph
- 11:21GraphRAG architecture
- 13:49Energy grid use case
- 17:20Agent architecture with graphs
- 19:08GraphRAG accuracy benefits
- 20:32Demo and GitHub
- 22:39Research on GraphRAG
- 24:44Learn more about GraphRAG
related talks
30mAgentic RAG with Knowledge Graphs Explained
36mAgentic GraphRAG: AI’s Logical Edge
30mGraph RAG: Bringing together graph and vector search to empower retrieval
24mUnlocking AI's Hidden Connections With Graphs
24mAgentic GraphRAG: AI’s Logical Edge
16mWhen Vectors Break Down: Graph-Based RAG for Dense Enterprise Knowledge
23mMeasuring the Performance of VectorRAG vs GraphRAG
33mVector Databases in the Age of AI: Embracing RAG for Enhanced Search Capabilities