Challenge
A UK dental group with 200 practices faced a documentation crisis:
- Time Drain: Dentists spent 10-15 minutes after each patient documenting findings
- Work-Life Balance: Many dentists stayed late to complete notes
- Inconsistency: Documentation quality varied widely between practitioners
- Privacy Mandate: Patient data absolutely could not leave the practice—no cloud services allowed
Traditional cloud-based transcription services were off the table due to strict data privacy requirements. The dental group needed AI that worked entirely within the browser, with zero external data transmission.
Solution
We built a 100% on-device AI documentation system that runs entirely in the browser:
How It Works
- Records Consultation: Audio captured during the patient appointment
- Local Transcription: Whisper model runs locally—no audio leaves the device
- Clinical Extraction: Gemma 3 via WebGL extracts findings, treatments, observations
- Template Population: Auto-fills the dentist’s own documentation templates
- Quick Review: Dentist reviews and corrects in under 2 minutes
Technical Innovation
The key breakthrough was running powerful AI models entirely in the browser using WebGL acceleration:
- Whisper for speech-to-text runs locally on the dentist’s computer
- Gemma 3 for clinical extraction runs via WebGL in the browser
- Zero network calls for any patient data
- Works offline once loaded
Privacy Architecture
[Microphone] → [Local Whisper] → [Local Gemma 3] → [Local Storage]
↓ ↓
(no cloud) (no cloud)Patient data never touches any external server. The practice owns and controls everything.
Results
The impact on dental practices was transformative:
| Metric | Before | After |
|---|---|---|
| Documentation time per patient | 10-15 mins | 2 mins |
| End-of-day admin backlog | 1-2 hours | Nearly zero |
| Documentation completeness | Variable | Consistent |
By the Numbers
- 85% reduction in documentation time
- 200 practices now live
- 0 patient data ever leaves the device
- GDPR fully compliant
Dentist Feedback
The reception from dental practitioners has been overwhelmingly positive:
- No more staying late to finish notes
- More time with patients, less time typing
- Consistent, complete documentation
- Peace of mind on data privacy
Technical Details
Tech Stack
- Speech-to-Text: Whisper (running locally)
- LLM: Gemma 3 via WebGL
- Runtime: Browser-based, no installation required
- Storage: Local browser storage, exportable to practice management systems
Why On-Device?
We chose on-device for several reasons:
- Privacy: Healthcare data is sensitive—local processing eliminates risk
- Compliance: GDPR compliance is simpler when data never leaves
- Reliability: Works even with poor internet connectivity
- Cost: No ongoing API costs for the dental practices
- Speed: No network latency for transcription
Integration
The system integrates with existing practice management software via:
- Copy/paste workflow (immediate)
- CSV export (batch)
- Direct API integration (for larger groups)
Project Details
- Duration: 2 months from concept to first practices live
- Team: 3 engineers (1 ML/WebGL specialist, 1 full-stack, 1 UX)
- Status: Live in 200 practices across the UK
- Compliance: GDPR compliant, zero cloud dependency
Need on-device AI for privacy-sensitive applications? Contact us to explore what’s possible.