Making Virtuous Cycles: Lessons From the Frontier of LLM-Knowledge Graph Interaction
summary
Describes concrete systems and results: a Neo4j-based Launch Guardian compares old and new sites (one example: a 60,000-page site) using vector indexing, BERT embeddings, KNN and Louvain to spot redirect and topical losses (one cluster loses 10,000 clicks), and a pitch tool that matches 500 brokers vs 25,000 agents by graph + cosine similarity. Notes a paper finding 40% noisy inputs, prescribes tight data models, label-based isolation, limited hop queries, agent-change auditing, and sample prompts/outputs used end-to-end.
AI-written summary · about
chapters (11)
- 0:00Introduction and team intro
- 1:56What we mean by virtuous cycles
- 3:22Case study: Launch Guardian project
- 8:22Lessons learned from Launch Guardian
- 11:54Clear goals before building
- 14:05Case study: Pitch tooling for outreach
- 15:54Describing knowledge graphs to agents
- 18:28User inputs and graph expansion
- 21:02Ontology expansion pipeline
- 22:25Demo: Graph queries and enrichment
- 29:21Summary and Q&A
related talks
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
23mKnowledge Graphs and LLMs in Action
01h 00mGraph Mastery to AI
28mEnhancing LLMs with Graph Technology
41mEnhancing LLMs with Graph Technology
22mBuilding Contextual Knowledge Graphs to Adapt, Evaluate, and Leverage LLMs
42mMaking OSS AI Real with Knowledge Graphs
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs