Building Contextual Knowledge Graphs to Adapt, Evaluate, and Leverage LLMs
summary
Combines Geonames, OpenStreetMap, Yelp, Wikidata and event/weather APIs into a Neo4j graph that adds temporal, situational and demographic attributes to places, updates hourly, and serves a REST endpoint to LLMs via Semantic Kernel; uses subgraphs, indexing and APO to scale Cypher over ≈45M nodes, verifies outputs by checking event+distance (example: walking-distance <2 km) and enforces zero hallucination by grounding facts, stores user intent vectors and a vector index for retrieval, and calls out Cypher as the main bottleneck and optimization target.
AI-written summary · about
chapters (12)
- 0:00Introduction and background
- 0:59Problem identification
- 2:13Contextual knowledge graphs
- 4:10Reasoning and graph design
- 6:14Graph retrieval in practice
- 7:14Data construction and flow
- 8:51Situational context and logic
- 10:20Evaluation and API strategy
- 11:05Schema and semantic design
- 12:36Implementation and optimization
- 16:48Graph demonstration
- 18:03Q&A and concluding remarks
related talks
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
30mBuilding Knowledge Graphs With LLMs
25mEnhance LLM Applications' Trustworthiness and Explainability
14mLLMs and Knowledge Graphs: A Case Study in Blue(y)
30mMaking Virtuous Cycles: Lessons From the Frontier of LLM-Knowledge Graph Interaction
30mMaking Virtuous Cycles: Lessons From the Frontier of LLM-Knowledge Graph Interaction
32mMulti-Agent Query Generation w/ LLMs on Complex Knowledge Graphs
44mGive Your LLMs a Left Brain