vancornelio.dev — boot sequencetty1
[POST ] cold boot
> _
[██░░░░░░░░░░░░░░░░░░░░░░] 9%
warming up… don't unplug

cold start · v2026.5

← All projects

MedBuddy

Work in progress

FlutterDartRiverpodSupabaseGoogle ML KitTFLite

Medication adherence app with on-device AI verification. Flutter + Riverpod + Supabase; Google ML Kit face detection and TFLite verify a dose offline, with local notifications and background scheduling for reminders.

Problem

Medication adherence fails quietly. Reminders get dismissed, and nobody, including the patient, knows for sure whether the dose was taken. MedBuddy turns the reminder into a short verification with the phone camera, works without a signal, and keeps the schedule running when the app is closed.

Decisions

  1. 1. Verify on device

    Chose: Google ML Kit face detection plus a TFLite model running on the phone.

    Rejected: Uploading frames to a cloud inference endpoint.

    Why: Health data stays on the device, verification works offline, and there is no per-call inference bill.

  2. 2. Background scheduling

    Chose: Workmanager tasks and local notifications drive the reminder schedule.

    Rejected: Server push as the only reminder channel.

    Why: Reminders must fire on time with no connectivity; the server is a sync target, not the clock.

Try it