EventKernel: Multi-Hop Reasoning and GraphRAG for AI-Powered Event Intelligence
summary
Claims a hybrid retrieval that beats vector-only by anchoring embeddings to a Neo4j domain graph and running deterministic traversals; the demo populates ~50 documents across 100 conferences and ~1,000 events and shows GraphRAG returning 243 nodes/266 relationships from five similar chunks versus sparse vector-only hits. Uses chunk embeddings, GraphRAG anchored in Neo4j, explicit bridge via presented-by user nodes, multi-hop DFS/BFS traversals, an agent router plus NL2Cypher retriever, and an evaluator agent; code is promised open source.
AI-written summary · about
chapters (10)
- 0:00Introduction to GraphRAG
- 4:23Domain graph for events
- 6:32Building lexical graph
- 9:40Vector RAG vs Graph RAG
- 11:26Neo4j graph visualization
- 15:03Connecting lexical and domain graphs
- 16:51Multi-hop reasoning in GraphRAG
- 18:40Demo comparison Vector vs Graph
- 23:40Agentic AI extension
- 26:28Q&A and closing
related talks
27mWhen Vectors Aren’t Enough: Unlocking AI’s Hidden Connections with Graphs
30mAgentic RAG with Knowledge Graphs Explained
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
16mEnhancing Retrieval-Augmented Generation with GraphRAG Patterns in Neo4j
01h 17mGraph Database and Retrieval Augmented Generation
36mAgentic GraphRAG: AI’s Logical Edge
30mAgentic RAG Meets Neo4j: Multi-Hop Reasoning Over Scientific Knowledge Graphs
19mBeyond Retrieval: Adding a Reasoning Layer to RAG