VoiceMed AI is our entry for the AssemblyAI Voice Agent Hackathon — a voice triage nurse we call Aria. The idea came from a simple observation: describing pain to a health system usually means filling forms or navigating phone menus, which is genuinely hard for elderly people, people with low vision, or anyone in pain. So we built someone you can just talk to. You press start, allow the mic, and speak. Aria asks one question at a time, in plain language, and you can interrupt her whenever you want — just like a real call. Under the hood there's a single WebSocket to the AssemblyAI Voice Agent API doing speech-to-text, LLM orchestration and text-to-speech, while our FastAPI server runs the actual clinical work: symptom lookup, drug-interaction checks, ESI 1–5 triage scoring, emergency escalation (call 911 / 988), appointment booking, and a SOAP note with ICD-10 codes that you can download at the end. We also built a dashboard so you can watch Aria think — a live transcript, a progress tracker that lights up as the interview moves through its stages, an ESI gauge, a medication safety panel, and a big red emergency banner when things get serious. For presenting, we added a little "Demo Copilot" that shows you what to say next and auto-advances when Aria hears you. One thing we cared about a lot: safety. Red-flag questions (chest pain, stroke signs, suicidal thoughts) are a mandatory part of the interview, the final urgency score is decided by deterministic local code — not the LLM — and Aria never diagnoses. It works in English and Spanish, has 44 automated tests behind it, and we verified the whole protocol live against AssemblyAI's real endpoint.
Category tags: