Clinics and small medical practices lose real time and money to phone-tag scheduling. This project shows a complete, working voice agent that handles the entire appointment-booking call end to end — no human needed for the scheduling itself. Built entirely on AssemblyAI's Voice Agent API, which combines speech-to-text, LLM reasoning, and text-to-speech in a single WebSocket connection. The agent greets the patient, asks for a preferred time, and calls a live scheduling tool to find real available slots. It negotiates naturally until the patient picks a time that works, repeats the full date, time, and clinic name back for confirmation, and only books the appointment after an explicit "yes." The scheduling backend is a FastAPI service deployed on Google Cloud Run, using SQLite with transactional booking (BEGIN IMMEDIATE) so two callers can never double-book the same slot. The agent's tools are wired directly to this live backend via AssemblyAI's HTTP tools feature. Built solo, combining 20 years of healthcare/regulatory domain experience with hands-on AI engineering — from idea to a fully working, deployed voice agent.
Category tags: