From Documents to Dynamic Graphs: Ontology-Driven Clinical Graph Construction
summary
An extended pipeline built on neo4j-graphrag converts clinical notes into a live knowledge graph by extracting LLM-identified entities, validating them against a local UMLS mirror, and merging by Concept Unique Identifier so identical concepts across documents join into one node. The system runs on-prem LLMs (CUDA GPUs), tracks provenance for every chunk and citation, flags mismatches for human review, and returns RAG answers (demo: lupus nephritis treatments) while noting UMLS misses SNOMED CT AU/ICD10-AM terms. UMLS-based normalization and provenance-validated merging are the core claims.
AI-written summary · about
chapters (10)
- 0:00Introduction and agenda
- 1:26Clinical knowledge challenges
- 3:34Digital health governance
- 4:33Use cases for graph RAG
- 8:01Graph RAG implementation
- 13:31UMLS and pipeline architecture
- 21:25Pipeline validation techniques
- 23:38Graph RAG data modeling
- 24:50Live demo and application
- 31:06Conclusion and resources
related talks
15mGraph-Driven Knowledge Retrieval: Neo4j for Healthcare Building Codes
28mGrounded Intelligence Boosting LLMs with Real-Time Graph Context
28mEnhancing LLMs with Graph Technology
20mPractical GraphRAG: Making LLMs smarter with Knowledge Graphs
30mBuilding Knowledge Graphs With LLMs
26mPractical GraphRAG Making LLMs smarter with Knowledge Graphs
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
30mNeurosymbolic Architecture in Production with Neo4j, GDS, Node2Vec & LLM Grounding