MedVoq
MedVoq is a voice AI for clinical documentation — it listens to a live doctor and patient consultation in any language and turns it into a structured, doctor-approved medical record.
The Problem
Doctors lose a large part of every visit to writing documentation instead of focusing on the patient. MedVoq gives that time back while keeping a human in final control.
What I Built
As full-stack and AI engineer, I built the entire AI system end to end.
- Real-time multilingual transcription of live consultations
- An LLM pipeline that extracts medications, dosages, and instructions into a strict structured schema
- A doctor-in-the-loop approval flow so nothing reaches a patient without review
- FastAPI backend and React frontend
The Hard Part
Reliability. Raw transcription of messy, accented, code-switched medical speech is noisy, and a single wrong drug name or dosage could be dangerous. I constrained the model to output a strict schema instead of free text and added verification, so the output stayed safe and trustworthy exactly where models tend to fail.
Outcome
Live in production.
Tech Stack
- React
- FastAPI
- Python
- LLMs
- ElevenLabs (live transcription)
- PostgreSQL
