Skip to content

Releases: hutchpd/AI-Medical-Scribe

AI Medical Scribe v1.1

19 Mar 10:10
@hutchpd hutchpd
V.1.1
8d90222

Choose a tag to compare

View all tags
AI Medical Scribe v1.1 Latest
Latest

AI Medical Scribe v1.1

This release introduces the biggest expansion of the prototype so far, adding structured clinical extraction, a confidence-aware review workflow, local audit logging, client-side FHIR export, optional direct FHIR delivery, and stronger local privacy controls.

Added

  • Structured extraction from transcript, notes, and summaries into clinically useful buckets
  • Review mode with confidence highlighting, provenance cues, stale/needs-review badges, and quick actions
  • Confidence indicators and low-confidence transcript filtering
  • Local append-only audit log with export to text or JSON
  • Client-side FHIR R4 document Bundle export
  • Optional browser-side POST of FHIR payloads to a configured endpoint
  • Structured Composition sections and optional clinical resource enrichment in FHIR exports
  • Optional encrypted session storage using Web Crypto
  • Passphrase unlock mode and session-only key mode
  • App lock/unlock controls with inactivity auto-lock
  • Ephemeral consultations, retention controls, purge-on-close, and stronger destructive deletion options

Improved

  • Session history now supports richer review and traceability workflows
  • Storage model clarified so FHIR exports are generated on demand and not stored separately by default
  • Privacy documentation expanded to reflect FHIR sending and local credential handling
  • Troubleshooting guidance now includes FHIR delivery issues
  • Architecture and roadmap documentation updated to reflect interoperability support

Notes

AI Medical Scribe remains a browser-based, local-first prototype with no project backend and no API keys. Data stays on-device by default, unless the user explicitly sends a FHIR payload to an external endpoint.

Assets 2
Loading

V.1.0

17 Mar 18:06
@hutchpd hutchpd
V.1.0
d8357e3

Choose a tag to compare

View all tags
V.1.0

AI Medical Scribe – Initial Prototype Release

This is the first public prototype of AI Medical Scribe, a browser-based, local-first tool for consultation transcription, summarisation, and document drafting.

Highlights

  • Live transcription using Chrome speech recognition
  • On-device AI summaries using Chrome Prompt API
  • Document drafting from transcript content
  • Session history with full local persistence
  • No backend – everything runs and stores data in the browser

Key Characteristics

  • Fully local-first: no server, no external AI services used by this project
  • Built on experimental Chrome built-in AI APIs (Gemini Nano)
  • Designed as a proof-of-concept for real-time consultation workflows

Important Notes

  • Requires a recent Chrome build (Canary recommended)
  • Requires enabling Chrome AI flags
  • First use may trigger a large on-device model download
  • Not supported on mobile browsers
  • Not suitable for clinical use

Known Limitations

  • Dependent on evolving Chrome experimental APIs
  • Performance depends on local hardware
  • Speech recognition behaviour varies by browser

Getting Started

See the README for setup instructions, including:

  • Running on localhost
  • Enabling Chrome flags
  • Verifying model availability

Purpose of This Release

This release is intended to:

  • Demonstrate feasibility of a fully local AI medical scribe
  • Explore Chrome’s on-device AI capabilities
  • Provide a base for further iteration and experimentation

Feedback, issues, and ideas are welcome.

Loading