G.A.M.E.R.S: Graph Agents with Multimodal Entities and Reasoning Schemas
Patient data lives in PDFs, radiology scans, lab results, and clinical notes. Most AI systems read each one alone, and context disappears between encounters.
G.A.M.E.R.S (Graph Agents with Multimodal Entities and Reasoning Schemas) fixes that with graph-based context memory. See how graph agents extract multimodal clinical entities, traverse disease hierarchies and comorbidity chains, and combine dense embeddings, BM25, and graph traversal in one hybrid retrieval pipeline built on FHIR and the WHOLE framework.
Walk away with the architectural patterns, live demos, and graph schemas to build agents that remember every patient encounter and reason across modalities, running on open-source models in a fully decentralized, private compute stack.
Guests: Krishnendu Dasgupta ( https://www.linkedin.com/in/krishdasgupta ) & Julia Hitzbleck ( https://www.linkedin.com/in/julia-hitzbleck/ )
GraphTalk Pharma: https://events.neo4j.com/graphtalkpharmalifesciencesde
Care4Rare: https://neo4j.com/customer-stories/care-for-rare/
Neo4j Agent Memory: https://neo4j.com/labs/agent-memory/
Chapters
- 00:00Introduction & Guest Welcome
- 02:04The Problem: Patient-to-Clinical Trial Matching
- 07:59Introducing GAMERS: Graph Agents Framework Overview
- 13:53System Architecture: Data, Models & Context Gateway
- 22:02Knowledge Graph Schema: FHIR, ICD-11 & Medical Ontologies
- 23:34Agent Memory & the Stateful Agent Pipeline
- 28:35Canonical Evidence & Deterministic Finalization Rules
- 33:03Live Demo: Patient Roster & Cohort Intelligence
- 41:07Live Demo: Trial Matching & Agent Reasoning in Action
- 54:23Q&A: Temporal Data & Why RLM Outperforms Standard LLMs
Related Videos
070 ESG Supply Chain Knowledge Graph – NODES2022 – Dr. Julian Gruemmer
Introducing Workspace — The Fastest Way To Build Your Neo4j Database
