Built by YGL · Health & Wellness
FitRas
FitRas is a wellness platform where AI does the work that makes nutrition tracking unbearable: identifying the food, estimating the portion, and calculating the macros. It runs on the web and ships as a native iOS and Android app from the same codebase.
In active development · web, iOS and Android builds
- AI integration points
- 7 distinct flows
- Specialised agents
- Coach · Nutritionist · Health
- Platforms
- Web · iOS · Android
- Recognition guardrail
- Portion gate before macros
The problem
Every nutrition app asks the user to do the hardest part themselves. Search a database, guess a portion, pick between nine versions of 'dal'. Most people give up in a week — not because they stopped caring, but because logging a meal costs more effort than the insight returns. The failure is worse for regional cuisines, where the food databases that Western apps rely on barely have coverage.
The vision
Take a photo. Confirm what it is. Get numbers you can trust. Everything else — coaching, workout plans, fasting, measurements — hangs off that one interaction being fast and honest enough to repeat daily.
AI architecture
How the intelligence works.
Food vision — the pipeline that refuses to guess
The most-used flow in the product, designed with reliability ahead of speed. A wrong calorie number that looks confident is worse than asking one clarifying question.
Capture
Capacitor · ML KitPhoto from the native camera, or a barcode scanned on-device.
Recognise
Gemini 2.5 Flash VisionVision model returns candidate foods with quantity, unit and a confidence score.
Portion gate
API-level enforcementIf the portion is ambiguous the API returns 422 needs_clarification and the user is asked. Macros are never invented.
Calculate
Genkit flowLocal food database checked first on the fast path; misses fall through to model calculation with regional cuisine knowledge.
Persist
Supabase PostgresMacros written to the daily log, streaks and FitScore recomputed.
What we built
The shipped surface area.
Everything below exists in the product's codebase. Nothing on this list is planned, aspirational or coming soon.
Two-step food vision pipeline: recognition, then nutrition calculation
A portion gate that returns HTTP 422 rather than guessing macros
RAG memory over pgvector so the AI coach remembers preferences and goals
Three specialised AI agents with separate prompts, memory and rate limits
Recipe URL intelligence — paste any recipe link, get per-serving macros
Predictive meal suggestions ranked by goal alignment
AI-personalised 7-day workout plans linked to an exercise video library
Daily FitScore card synthesising macros, workouts, hydration and sleep
Intermittent fasting tracker with 16:8, 18:6, 20:4, OMAD and custom protocols
Body measurement tracking across six sites with trend charts
On-device barcode scanning for packaged foods via ML Kit
Chrome extension for capturing recipes from the browser
Push notification system with scheduled meal reminders
Phone and email OTP authentication with rate limiting
Product experience
What it feels like to use.
Log a meal in one interaction
Camera, confirm, done. The portion dialog only appears when the model is genuinely unsure, so the common path stays a single tap and the uncommon path stays accurate.
Three coaches, not one chatbot
An AI Coach for training and motivation, an AI Nutritionist for meal planning and macros, and an AI Health Assistant for sleep, stress and hydration. Each has its own system prompt and its own memory namespace, so advice stays in its lane.
A dashboard that loads once
Daily totals, streak, activity chart and health advice arrive from a single consolidated endpoint with short-TTL caching and request deduplication — replacing the several separate calls the first version fired on every load.
Native where it matters
The same codebase ships as web, iOS and Android. Camera, haptics, barcode scanning and push notifications use native capabilities rather than web approximations.
Technical depth
The decisions that don't show up in a screenshot.
These are the parts that separate a demo from a product — and the parts most worth talking about with a technical buyer.
Quality gates in CI, not vibes
The nutrition pipeline is evaluated on every build against a fixed dataset. Recognition success rate must stay at or above 90% and the wrong-accept rate at or below 2%, or the build fails. Portion-gate accuracy is monitored with SLO alerting.
RAG memory over pgvector
Conversations write durable memories — preferences, habits, goals, health constraints — as embeddings. Each new prompt retrieves the top matches by cosine similarity and composes them with the user profile, recent history and today's progress.
Model fallback routing
Gemini 2.5 Flash is primary, with OpenRouter and HuggingFace-hosted open models configured as fallbacks. A single provider's outage degrades quality rather than taking the feature down.
Security-critical feature gating
Paid features resolve through one manifest and a single unified flag, so entitlements unlock atomically instead of drifting between screens. Server-side tier checks sit on the endpoints themselves, not only in the UI.
Row-level security by default
Postgres RLS policies scope every user-owned table, so a missing filter in application code cannot leak another user's meal log.
Stack
What it's made of.
Product
- Next.js 16
- React
- TypeScript
- Tailwind CSS
- Radix UI
AI
- Google Genkit
- Gemini 2.5 Flash
- pgvector
- OpenRouter
- ML Kit
Platform
- Supabase
- PostgreSQL
- Redis
- Firebase Cloud Messaging
Mobile
- Capacitor 8
- iOS
- Android
- RevenueCat
Operations
- Vercel
- Sentry
- Playwright
- Vitest
Current state
Where the product is today.
Stated plainly, including what hasn't happened yet.
Web application builds and deploys; Android release artefacts and an iOS build have both been produced.
Nutrition pipeline runs behind automated evaluation with enforced accuracy thresholds.
Barcode scanning for packaged foods is the feature currently in progress.
Launching within the week — no adoption or revenue figures are claimed anywhere on this site.
Work with us
Need this kind of engineering?
The practices on this page are the ones we bring to client work.