Healthcare navigation is broken. In Italy alone, 5.8 million people gave up on medical visits in 2024 due to waiting lists or costs, nearly 10% of the population. Three in ten hospitalizations could have been avoided with better routing to territorial services. And companies watch only 5% of corporate welfare budgets get used for health and prevention.
The gap isn't a lack of services. It's that people can't find the right one. They describe symptoms and needs in natural language; the system offers them a directory.
Geen.ai is an AI triage and orientation layer that interprets natural language and guides each person to the most appropriate health service, which means less wasted time, fewer dead-end attempts, and lower costs. For companies, health benefits become real value through targeted utilization. For healthcare organizations, booking requests become more appropriate access with less load on clinical staff.
Triage AI in healthcare is a guardrails-first problem: the system must understand messy human language about symptoms and needs, route with high confidence, and know exactly when to hand off to a human. We worked on the AI layer's engineering: the natural-language understanding and routing infrastructure where a wrong answer isn't a bad UX moment but a health decision.
Every request runs through the same guardrails-first pipeline: the LLM interprets messy natural language, a GraphRAG knowledge graph routes to the right subspecialty, then deterministic filters rank real, reachable specialists — no diagnosis, no PII stored.
User describes symptoms in natural language via the chat widget, with sex and age.
Backend searches the knowledge base for rules matching symptom, gender, and age band.
The LLM returns primary and secondary specialty, urgency, red flags, and gender-medicine notes.
Neo4j traverses Symptom → Subspecialty → Physician using knowledge extracted from PubMed.
Filter physicians by tenant coverage — SSN, UniSalute, Metasalute, FASI, and more.
Rank by geographic distance from the patient's coordinates.
Urgent symptoms escalate and alert the project owner.
The person is guided to the most appropriate service — anonymized log stored, no diagnosis, no PII.