Public Healthcare: Either a Labyrinth or a Graph — GraphRAG Routing on Neo4j

Session track: AI Engineering

Session time: 08:00 am - 08:30 am UTC November 12

Session description:

One in ten Italians went without needed care in 2024 — often not for lack of a doctor, but because the system's knowledge is scattered across tables, PDFs, and regional portals: a labyrinth only experts can navigate. In this session, Giulia Marchese will show how Geen.ai turns that labyrinth into a graph, using a production Neo4j knowledge graph that an AI orientation layer traverses to route a citizen — from a free-text symptom to the right specialist, facility, or social service — in seconds. She will focus on the "how": the data model that encodes Italy's regulatory care pathways as one graph with clinical, provider, and social layers; the multi-hop Cypher that powers the routing; and how a guideline "red flag" short-circuits a specialist match into an urgent pathway. She will cover the change-data-capture pipeline that keeps the graph trustworthy, the human-in-the-loop governance on every graph write, and a privacy-by-design model that stores relationships between symptoms and services, not between data and people. You will leave with concrete patterns for building a knowledge graph in a regulated domain — and an honest account of what was hard: separating a reasoning engine from a routing engine, moving from string matching to graph-native full-text and vector resolution, and treating schema as code.