LLMs and Knowledge Graphs: A Case Study in Blue(y)
summary
A working Bluey proof‑of‑concept builds a Neo4j knowledge graph of characters, episodes and extracted concepts, uses an LLM to extract three concept types per recap (title, description, quotes), and indexes those concept texts as vectors. Queries translate to Cypher for deterministic family/appearance answers and to vector search over concept nodes for thematic queries, then traverse back to recaps for context. Combining graph + vectors reduces hallucinations and returns more complete, traceable answers.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:09What is RAG and why it’s not enough
- 3:58Limitations of vector-based RAG
- 5:22What are knowledge graphs
- 7:12Integrating graphs into RAG
- 9:22Graph design for characters and episodes
- 10:51Questions about series concepts
- 13:24Summary and demo
related talks
23mKnowledge Graphs and LLMs in Action
44mGive Your LLMs a Left Brain
25mEnhance LLM Applications' Trustworthiness and Explainability
30mBuilding Knowledge Graphs With LLMs
22mBuilding Contextual Knowledge Graphs to Adapt, Evaluate, and Leverage LLMs
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
29mPowering Legal Research with Knowledge Graphs and Large Language Models
32mMulti-Agent Query Generation w/ LLMs on Complex Knowledge Graphs