Enhance LLM Applications' Trustworthiness and Explainability
summary
Claims that combining Diffbot’s web-scale knowledge graph with Neo4j produces transparent, verifiable LLM answers by linking extracted entities and source articles into a Neo4j DB and exposing provenance for each fact. Implements a joint demo that ingests news, runs entity and relation extraction via Diffbot APIs, stores nodes and embeddings in Neo4j, and answers queries (e.g., a long supplier list for NVIDIA) using vector + graph search and entity linking. Works live but depends on initial KG quality and ongoing human-in-the-loop validation.
AI-written summary · about
chapters (9)
- 0:00Introduction to trustworthy LLM apps
- 1:36What are knowledge graphs
- 3:43Why use knowledge graphs for LLMs
- 5:52Demo of graph-based LLM application
- 10:12How to build quality knowledge graphs
- 14:07Limitations of LLM entity extraction
- 18:46Live entity extraction demo
- 22:22Vector vs graph databases for LLMs
- 24:08Q&A and closing remarks
related talks
41mEnhancing LLMs with Graph Technology
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
44mGive Your LLMs a Left Brain
23mKnowledge Graphs and LLMs in Action
28mEnhancing LLMs with Graph Technology
32mTruth-seeker: Using LLM agents to build and verify knowledge bases
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
52mGive Your LLMs a Left Brain