Skip to content

Graph-based Healthcare Series

This ongoing series explores how graph representations can power intelligent clinical tools—covering everything from knowledge modeling to real-time decision support with large language models (LLMs).

đź“Ś Posts in this Series

  1. Modeling Medical Guidelines as Interactive Graphs Explore how we transformed IMNCI guidelines from dense clinical tables into a structured, interactive Neo4j graph, enabling machine-readable logic and unlocking new possibilities for AI-powered healthcare tools.

  2. Enhancing Patient Diagnosis with Graph Based Retrieval Augmented Generation Building on our graph model, this post introduces a hybrid graph RAG system that combines graph traversal with LLMs to deliver grounded, context-aware diagnostic support—even from partial clinical input.

  3. Managing Agentic Flows with Pydantic Graph Deep dive into how we designed an agentic, step-by-step diagnostic assistant using pydantic graph—enabling LLMs to drive interactive clinical sessions that adapt to patient-specific conditions, resource constraints, and evolving graph state.

  4. Simulating Real World Pediatric Encounters Using Large Language Models Learn how we use LLMs to generate, verify, and structure synthetic pediatric encounters based on IMNCI logic—laying the foundation for Bayesian diagnostic modeling in uncertain, real-world conditions.

  5. Bayesian Pattern Recognition For Real World Applications In this post, we build a Bayesian diagnostic engine from scratch using IMNCI-verified synthetic patient data. By transforming clinical observations into atomic features and computing log-likelihood ratios, we can construct interpretable prototypes for each condition, score patients based on real evidence, and update beliefs using Bayes’ rule. This allows us to mirror how clinicians think under uncertainty.

  6. Evaluating GraphRAG vs. RAG on Real-World Messages We benchmark GraphRAG and a traditional RAG pipeline on real pediatric triage messages collected from Last Mile Health, showing how graph-structured retrieval leads to higher recall, fewer hallucinations, and more trustworthy next-step clinical decisions.

  7. Longitudinal Patient Care Coming soon!

  8. Detecting Emerging Trends in Patient Healthcare Coming soon!

More entries to follow-stay tuned!