Adaptive Knowledge Graphs for Dynamic RAG Systems
summary
Shows a working GraphRAG pipeline that combines embeddings for semantic search with Neo4j graph traversal to answer multihop queries (e.g., "which ex‑Google founders recently raised funding") and keeps facts live via streaming ingestion. Uses an LLM-powered KG builder to extract entities and relationships, a Kafka CDC→Neo4j sync for real‑time updates, Cypher for traversal, vector search for documents, and relationship confidence scores with a confidence threshold 0.9 / review 0.7–0.9 plus 10–20 edges/day for HITL to close the loop.
AI-written summary · about
chapters (9)
- 0:00Introduction to adaptive KGs
- 3:15Knowledge graphs vs vector databases
- 8:55Biomedical adaptive KG example
- 13:16Why Neo4j for adaptive KGs
- 14:01Real-time updates with Neo4j
- 16:55LLM-powered graph building
- 18:44Human-in-the-loop oversight
- 25:57Living learning trustworthy systems
- 26:20Closing and future course
related talks
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
12mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
32mGraphRAG: Powering Up LLMs with Knowledge Graphs
25mEnhancing Job Matching With Knowledge Graphs and RAG
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
01h 40mKnowledge Graphs & GraphRAG: Techniques for Building Effective GenAI Applications
21mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
30mAgentic RAG with Knowledge Graphs Explained