Record or upload audio. Whisper transcribes it. MedGemma generates a structured clinical note. Both run locally on a physician's MacBook Pro. No cloud. No API. No data stored.
NOT MEDICAL ADVICE: This is a research and education demonstration. Neither Whisper nor MedGemma is validated for clinical use. Do not use this for medical advice, diagnosis, or treatment decisions. Always consult qualified healthcare professionals.
This tool may produce inaccurate transcriptions and/or inaccurate, incomplete, or misleading medical information. Outputs should not be relied upon for any health-related decisions.
This is an independent educational demonstration by Chief Health AI (PBC). It is not endorsed by, affiliated with, or approved by Google or OpenAI.
Audio is transmitted to a local machine for processing, transcribed, and immediately discarded. No audio, transcripts, or notes are stored, logged, or transmitted to any third party.
Audio is transmitted to a local machine for processing, transcribed, and immediately discarded. No audio, transcripts, or notes are stored, logged, or transmitted to any third party.
Two open-source models — OpenAI's Whisper and Google's MedGemma — running on a single MacBook Pro. Voice in, clinical note out. No vendor lock-in, no cloud dependency, no data leaving the machine.
Any AI can summarize what was said. The question is whether it flags what WASN'T said. The Information Gaps section tests whether MedGemma can identify missing clinical information — a capability where it scored only 7.2% on the HealthBench benchmark.
Built on OpenAI Whisper (open-source) and MedGemma by Google (open-source)