Healthcare AI Triage Assistant
A US senior-care network was triaging incoming patient calls the way most clinics still do - a human at a front desk, an Excel sheet, and a printed escalation tree on the wall. Urgent cases sometimes waited; routine cases sometimes got prioritised. They wanted an AI assistant that could listen to the symptoms a patient described in plain English, ask a few clarifying follow-ups, score urgency against the clinic''s own protocol, and surface a recommended next step (book GP, refer to specialist, send to ED) to the front-desk nurse for confirmation. We built it on a fine-tuned Llama-2 reasoning layer wrapped in a strict guardrail policy: the model never tells the patient a diagnosis, never overrides a clinician, and always defers to a human on the borderline cases. Every interaction is logged with the model''s confidence so the medical director can audit, retrain, and tighten the protocol.
Open case