Hi all, I’m Berk. With a small team I built Kalomi, a calorie, diet and workout tracker for Android, in Flutter.
Why: when I started counting calories to lose weight, I got fed up with the apps I tried. I’m Turkish, and everyday food like lentil soup, mantı or börek was either missing or had portions that made no sense. Then I read hundreds of calorie app reviews and the same complaint kept coming up: the same meal shows different calories on different days. I use Kalomi every day myself now.
The design decision: keep the model out of the arithmetic. The vision model only returns food + portion; calories and macros come from a verified food catalog (USDA data plus a hand-built Turkish cuisine catalog). Same meal, same numbers. If nothing matches, the model’s value is shown and marked as an estimate.
Flutter / backend bits:
- The catalog ships as a bundled SQLite file, copied to app storage only when its version stamp changes, so catalog search works offline.
- Photos are re-encoded before upload, which strips EXIF/GPS.
- Health Connect access is read-only.
- Supabase (Postgres + RLS) with Edge Functions in Deno between the app and the AI provider, so no AI keys live in the app.
Free: barcode scanning, manual entry, water, weight, history, PDF reports. Premium: photo analysis, AI coach, recipes, workout plans (2-day trial, no card, no auto-charge). Android 9+, English and Turkish, sign-in with Google or Apple.
Feedback welcome, especially on the photo flow and onboarding. Happy to go deeper on any of the implementation details.

