MediVoice — Evidence-First Patient Intake

MediVoice is an evidence-first patient intake voice agent designed to turn natural patient speech into a structured, clinician-facing handoff. Instead of asking patients to complete another long form or follow a rigid question tree, MediVoice lets them describe their symptoms naturally in one pass. The browser captures microphone audio and streams it to AssemblyAI Universal-3.5 Pro Realtime using the v3 Streaming API with Medical Mode. As the patient speaks, MediVoice displays the live transcript and structures key intake information including chief complaint, duration, pain level, location, symptoms, medications, and allergies. When the recording ends, it generates a concise clinical note and a structured doctor intake report. MediVoice is deliberately designed as a support layer, not an autonomous clinical decision-maker. The original transcript remains visible so clinicians can review what the patient actually said. The system does not diagnose, prescribe, or replace clinical judgment. The permanent AssemblyAI API key never reaches the browser. A small Node.js backend creates a short-lived streaming token for the live session. The full microphone-to-transcript-to-structured-report workflow has been tested end to end with a real AssemblyAI API key. MediVoice’s goal is simple: reduce intake friction while keeping clinicians firmly in control.