AI Medical Scribe is an experimental browser-based prototype for local-first medical transcription and documentation workflows.

It explores what an ambient scribe might look like if it lived entirely in the browser, with no project backend, no API keys, and no project-side data leaving the device by default. The prototype uses Chrome’s built-in speech recognition and on-device AI features for live transcription, summarisation, document drafting, structured extraction, review support, local audit logging, and client-side FHIR export.

This is a technical experiment rather than a production clinical tool. It is designed to test whether a browser-based, local-first approach could improve privacy, trust, and workflow fit compared with cloud-dependent scribe tools.

What it does

• Live transcription during a consultation using Chrome speech recognition
• Manual note capture alongside the transcript
• Important-moment markers within the transcript timeline
• On-device summary generation after a session ends
• Rich text document drafting from transcript content using configurable templates
• Structured extraction into clinically useful buckets such as problems, medications, allergies, investigations, follow-up actions, diagnoses, safety netting, and admin tasks
• Review mode with confidence highlighting, provenance cues, and quick validation actions
• Local append-only audit logging for traceability
• Client-side FHIR R4 document Bundle export
• Optional direct browser-side FHIR delivery to a configured endpoint
• Optional encrypted local session storage, app lock controls, retention settings, and destructive local deletion workflows

How it works

AI Medical Scribe runs as a local-first front end in the browser.

• Transcription uses Chrome speech recognition when available
• Summaries and document drafting use Chrome’s built-in on-device AI capabilities when available
• Structured extraction turns transcript, notes, and summary content into reusable buckets for review and downstream export
• Audit logging records important user and system actions locally in the browser
• FHIR export generates structured FHIR R4 JSON on the client side, with optional direct POST to a configured endpoint

No project backend is required.

Privacy and security approach

The project is built around a local-first model.

• Transcript, notes, summaries, documents, and FHIR export data are not sent to any backend controlled by the project
• Session data is handled in-browser and can optionally be encrypted at rest
• App-level locking, inactivity locking, retention settings, purge controls, and ephemeral consultation mode are available
• If a user explicitly configures an external FHIR endpoint and chooses to send data, that payload is sent directly from the browser to that endpoint

This project does not claim to solve governance, compliance, or deployment requirements on its own. It is an experiment intended to explore what a more privacy-conscious browser-based scribe could look like.

Important limitations

• Depends on Chrome-specific features that are still evolving
• Requires relatively modern hardware for on-device AI features
• Speech recognition quality depends on browser behaviour and environment
• Not suitable for clinical use
• Not a medical device

AI Medical Scribe is best understood as an open source prototype that explores whether local-first, browser-based medical scribing is now practical enough to be worth taking seriously.