◈ AI4GOOD
•ai4good / showcase

NeoFit: Science-Informed Indian Health & Fitness in 22 Languages

Sehat, simple banayi. NeoFit is a science-informed Indian health app in 22 Indian languages — calorie tracking that knows dosa and biryani, on-device food recognition, AI food photos and exercise videos, powered by Azure AI.

ai4goodandroidkotlinazure-aihealthindia

Part of the #AI4Good series — one app a day, each built for good.

NeoFit — Sehat, simple banayi

Open almost any fitness app and it assumes two things: you eat like the West, and you read English fluently. For most of India, both assumptions are wrong. NeoFit is my answer.

What it is

NeoFit — “Sehat, simple banayi” — is a science-informed health app tailored to Indian dietary habits, offline-first and honest about estimation. What’s shipping (now v1.4):

  • Indian food intelligence — a seeded knowledge base spanning regions (South, North, North-East, West, Central, East, pan-Indian): home meals, street food, thalis, beverages, snacks and sweets — each with native names and per-serving nutrition.
  • Honest calorie estimation — portion multipliers + cooking-style adjustment + a confidence score (High / Medium / Rough estimate) and a human-readable basis; approximate values are prefixed with ~ and you can correct anything.
  • AI Coach — a Coach tab that answers questions grounded in today’s calories, macros and goals, and works offline.
  • 1,300+ exercise library with step-by-step form and demo videos filmed in a consistent NeoFit gym, plus build-your-own workouts, a live timer, voice (TTS) guidance, and on-demand AI-generated pose images.
  • Health Connect sync (steps / distance / active-calories) with a graceful estimate/simulate fallback, a transparent wellness score (0–100) you can see the breakdown of, and smart nudges.
  • Fully localized into 22 Indian languages — every screen, with native dish names and Hinglish/Kanglish flavour.

How it was built

NeoFit is the most architecturally serious app in the series — Kotlin + Jetpack Compose (Material 3), Clean Architecture with strict unidirectional flow: UI → ViewModel (StateFlow) → UseCase → Repository. The domain is pure Kotlin (no framework deps), and the parts that must be correct are isolated as pure, deterministic, unit-tested engines: CalorieMath, CalorieEstimationEngine, RegionClassifier, WellnessScoreEngine, RecommendationEngine.

  • Room is the single source of truth — everything works offline; network and AI sit behind interfaces with mocks.
  • Integrations are all pluggable ports: HealthConnectManager, FoodRecognitionService (a mock today, swappable for a real recogniser), and the food-image system that degrades gracefully web image → Azure-generated (gpt-image) → placeholder via FoodImageProvider. Azure AI is wired in behind ImageGenerationService / AzureImageGenerationService.
  • Stack: Room · Hilt · Coroutines + Flow · DataStore · Retrofit/OkHttp + kotlinx.serialization · Coil · Health Connect · JUnit + Truth + Compose UI Test. compileSdk 35, minSdk 26.
  • A deliberate call: one Gradle module with strict package separation (core, data, domain, engine, integration, feature/*, di) — it “compiles in one shot,” with the boundaries already drawn so it can split into modules later mechanically.

The theme throughout is honesty: the app tells you when a number is a rough estimate, and never pretends an approximation is a fact.

The good

Health advice only works if you can understand it and see your own food in it. By speaking 22 languages, knowing the Indian plate, and being transparent about every estimate, NeoFit meets more than a billion people where they actually are. Inclusion is a health outcome — and that’s #AI4Good.

Try it

Previous → KidKat · Next → GpsCamera. #AI4Good