Context
MedicAl Wallet gives patients one secure, patient owned health record connected to doctors, pharmacies and laboratories, on iOS, Android and the web.
Its core promise depends on one hard problem: reading medical documents correctly. The original OCR centric flow broke on every new layout.
The challenge
- 01
Unpredictable documents
Every laboratory formats reports differently, and patients upload photos as often as PDFs.
- 02
Trust
A wrong value in a health record is worse than no value. Uncertainty has to be visible.
- 03
Cost at scale
AI per document costs must stay profitable as the user base grows.
- 04
Privacy
Medical data requires GDPR compliance, anonymisation and strict access control.
Our approach
- 1
AI Vision as the primary extractor
We replaced brittle OCR flows with AI Vision extraction, returning structured, schema validated output with a confidence score for every value.
- 2
Uncertainty handled explicitly
Low confidence extractions are flagged for review instead of being silently saved, and name mismatches between report and account are detected during processing.
- 3
One canonical parameter dictionary
A multilingual dictionary unifies values that labs report under different names, languages or spellings, so charts and trends stay reliable.
- 4
Hybrid routing for growth
Documents are routed between OCR and AI Vision by document type, confidence and cost thresholds. Stable, well known lab layouts move to OCR as volume grows, keeping cost per document low without losing quality.
- 5
Security and quality
Personal data is anonymised before it reaches any AI model, identifiable data is separated from processing pipelines, and 75%+ test coverage gates every release in CI/CD.
Architecture
Personal data is anonymised before cloud processing. Role based access, ownership checks and audit trails protect every record.
Results in detail
| Phase | Delivered |
|---|---|
| 1. AI driven enhancement | AI Vision migration, schema validation, confidence scoring, cost monitoring, hybrid routing design, 75% test coverage, GDPR and anonymisation, security hardening |
| 2. Platform expansion | Doctor's Portal from the ground up, dashboard refactor with real time global search, full Albanian localisation, pharmacy and laboratory marketplace |
| 3. Data and clinical features | Database refactor separating analyses from parameter history, categorised doctor chat, Parameters screen with historical charts, enforced app version upgrades |
| 4. Platform improvements | Grouped notifications, free tier upload limits, duplicate account detection, name mismatch detection, improved charts and admin views |
