Lab project · AI · Health

SimpleCarb

Photograph your meal. Get a carb estimate you can trust. Log what you actually ate — and watch how your glucose responds.

iOS · React Native Claude Vision Dexcom CGM Local-first · Private

Three steps. One tap each.

Carb counting is the hardest daily task for someone living with type 1 diabetes. SimpleCarb makes it fast, accurate, and automatic — without ever telling you what to inject.

📸
01

Photograph the plate

Rest your thumb at the plate edge as a scale reference. One photo, full plate. The AI uses the thumb to estimate volumes — no weight scales, no ingredient scanning.

🤖
02

Get the estimate

Claude vision identifies every food, estimates carbs, protein, fat and calories, and flags its confidence. Your API key never leaves the server — only a number comes back to your phone.

🍽️
03

Log what you ate

Photograph leftovers, or tap All / Most / Half / Barely. The app stores both the AI estimate and your actual portion, so accuracy improves over time and nothing is lost.


Before & after.

Most apps estimate the full plate, then guess how much you ate. SimpleCarb photographs both: the meal as it arrives, and the leftovers when you're done. That second photo turns a rough estimate into a precise record.

🍝
Before 84 g carbs
🍽️
After 63 g eaten

Built around real use.

📡

Dexcom CGM integration

Live glucose from your Dexcom appears alongside every meal log. Correlate what you ate against what your glucose did — the insight that drives better decisions over time.

🔒

Private by design

All data lives on your device. No account, no cloud database, no third-party sync. The AI call passes through a proxy that adds the API key server-side and returns only the estimate — your photos are never stored anywhere else.

💉

Insulin & event logging

Log fast and long-acting insulin doses, hypo treatments, and life events alongside meals. The timeline shows everything together so patterns become visible.

One-handed, interrupted

Designed for real life: logging with children present, at a busy table, mid-conversation. Fewer taps always wins. The meal-in-progress banner rescues anything left half-done.


The stack.

Client Expo / React Native — iOS, local-first, SQLite on device
AI vision Anthropic Claude — meal photo → carb estimate via vision API
Proxy Cloudflare Workers — stateless, key stays server-side, never in the app
CGM Dexcom API — OAuth2, glucose history correlated with meal logs
Storage SQLite on-device — fully offline, no external database, no accounts
Status Active · in daily use — personal build, Phase 1 complete

Have an idea worth making real?

SimpleCarb is a SimplySoft lab project. If you're building something at the intersection of AI and health, let's talk.

hello@simplysoft.cz