Web
AortaLink
An open-source personal electronic health record for hypertension, built on HL7 FHIR R4, with machine learning that runs on the device.
Problem
People with hypertension need to record their blood pressure regularly and understand the pattern behind it, not just the daily number, without handing their health data to a third-party service.
Approach
I built AortaLink as an open-source personal electronic health record under deliberately strict rules: no mock data, no external AI services, and every number on screen must be traceable to data the user actually entered. The data is modelled on the HL7 FHIR R4 standard (Patient, Observation with LOINC codes, MedicationRequest) so it can be exchanged with other health systems.
Solution
- On-device machine learning, written in plain TypeScript with no network calls: trend forecasting with OLS regression and a 7-day prediction interval, detection of patterns such as the white-coat effect and morning surge, and a medication adherence model using logistic regression trained on the device.
- A local assistant whose every answer cites the data it came from, with no generative AI.
- Bluetooth blood pressure monitors read straight from the browser through Web Bluetooth (Blood Pressure Profile).
- Offline-first with IndexedDB (Dexie), then sync to a server (Express + MongoDB) with per-record conflict resolution and deletion markers, so edits on two devices never overwrite each other.
- Circadian rhythm classification (dipper, non-dipper, riser) and measurement context (home, clinic, after medication).
- Encrypted backups using AES-256-GCM with the user's password, plus clinical PDF reports.