← All work
Project · Active ·2026

Turbine

A local-first, privacy-first clinical dictation and scribe app. Speech-to-text and note cleanup run on the provider's own device, and the app adapts to each clinician's vocabulary and style — without sending patient audio to the cloud.

Healthcare Clinical AI Privacy-first On-device

Turbine turns a clinician’s speech into clean, structured notes — entirely on their own device. The conviction behind it: the fastest way to earn a clinician’s trust with AI is to never let patient data leave their machine.

What it is

A native macOS and iOS app (with a browser reference build) that turns a clinician’s speech into clean, structured notes. Press a hotkey, speak, and the raw transcript appears instantly; an optional cleanup pass refines it into clinical prose — all running locally.

Local-first by design

  • Speech-to-text runs on-device (Whisper / WhisperKit / Apple Speech) — no audio leaves the machine by default.
  • Note cleanup uses a local language model (Ollama or Apple’s on-device models), with a deterministic fallback so the app never blocks on a network call.
  • Cloud transcription is strictly opt-in; structured outputs are review-only and never written into a record without the clinician’s approval.

Adapts to the provider

Turbine learns a clinician’s style through customizable personas, clinical vocabularies and shorthand, and reusable prompt templates — so notes come out sounding like the provider, not a generic model.

Why it matters

Turbine brings agentic technology to healthcare on the clinician’s terms — respecting privacy, trust, and a clinician’s time rather than ignoring them, on a HIPAA-conscious, privacy-first foundation.