WorkAI-Native Mobile · iOS & Android2026
An Arabic-first AI calorie tracker for iOS and Android: photograph your plate, and a vision model resolves it against a hand-curated Saudi dish database.
Independent product by our founding team · in active development
iOS + Android
Platforms
Arabic RTL
First language
86
Tests green
0
Keys needed to demo
Flagship mobile build iOS & Android in development
Sahtain صحتين — point your camera, get honest calories
An Arabic-first AI calorie tracker built by our founders. The vision model is a commodity — the curated truth it answers from is the product.
Snap a plate, get honest macros
A vision model reads the photo; a hand-curated Saudi dish database turns its guess into a defensible calorie range — kabsa, mandi, shawarma.
Arabic-first, RTL-native
Designed right-to-left from the first screen, with Arabic dish search — not an English app with translations bolted on.
Metered AI, honest billing
Free scans are quota-limited, the price shows before signup, and the quota's race condition is proven safe on a real database.
The challenge
Generic photo-calorie apps guess Western food and guess badly at Khaleeji cuisine. How do you make camera-based tracking honest for a Saudi audience — in Arabic, right-to-left, at a price shown before signup?
What we built
The insight the product is built on: the vision model is a commodity — the curated database its guesses get resolved against is the product. A cross-platform Expo/React Native app on a typed Hono backend, where a photo is classified by a vision model and then resolved against a hand-curated Saudi/Khaleeji dish database with per-dish calorie and protein ranges, so the answer is a defensible range from curated data rather than a hallucinated point estimate. Arabic is the first language, not a translation: RTL-native screens, Arabic dish search backed by trigram matching, English as the fallback. The engineering spine is proven end to end — contract-first API, generated typed client, real Postgres, store compliance gates — with the free-tier scan quota's race condition tested on the real database, because a metered AI feature that miscounts is a billing bug.
Key engineering
The decisions that made it work
Camera → vision model → curated truth
The model proposes, the database disposes: every AI guess resolves against curated Saudi dishes, and the app shows honest ranges, not fake precision.
Arabic-first, RTL-native
Designed right-to-left from the first screen, with Arabic dish search — not an English app with translations bolted on.
Metered AI, engineered honestly
Free photo scans are quota-limited; the quota's race condition is proven safe on real Postgres under concurrent scans.
Store-ready from day one
Compliance gates, data-safety declarations and release scaffolding built alongside the product, not after it.
The results
What actually changed
- 1
Full engineering spine proven end to end: contract → typed client → API → Postgres → Arabic RTL screens
- 2
86 automated tests green, including the scan-quota race on a real database
- 3
Runs completely with zero third-party keys in fake mode — the whole product demos without an OpenAI or billing account
What it doesn't do
Sahtain is in active development: the walking skeleton is proven and the screens shown are real, but the full 12-screen product — auth, diary, barcode, paywall — is still being built toward a Saudi iOS App Store launch. We show it because building in the open, limitations included, is how we work.
Stack
- React Native
- Expo
- TypeScript
- Hono
- PostgreSQL
- pg_trgm
- Vision models
- RevenueCat
- EAS
- Fastlane
Tell us the pain point. We'll tell you honestly what AI can do about it.
A founder replies within 24 hours. If the answer is 'AI is wrong for this', you'll hear that too — free either way.