Making Virtuous Cycles: Lessons From the Frontier of LLM-Knowledge Graph Interaction
summary
Project Launch Guardian loads before-and-after site crawls into Neo4j (about 60,000 pages) and uses vector indexing plus cosine similarity to detect mismatched redirects that cost as much as lost 10,000 clicks on a subtopic. Topic modeling uses BERT + KNN + Louvain to cluster terms, LLM agents generate page-level recommendations, and on-demand knowledge graphs spin up with label-based isolation, hop limits, and change trails so agents can alter the data model while leaving an audit.
AI-written summary · about
chapters (9)
- 0:00Introduction and team
- 1:56Virtuous cycles with LLM and graphs
- 3:22Case study Launch Guardian
- 8:22Lessons learned from Launch Guardian
- 14:05Pitch tooling project overview
- 15:54Describing knowledge graphs to agents
- 18:28User inputs and graph expansion
- 22:25Data model and enrichment
- 29:21Summary and Q&A
related talks
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs
23mKnowledge Graphs and LLMs in Action
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
22mBuilding Contextual Knowledge Graphs to Adapt, Evaluate, and Leverage LLMs
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
28mEnhancing LLMs with Graph Technology
41mEnhancing LLMs with Graph Technology