CaloGlow is a mobile-first Progressive Web App for calorie and nutrition tracking, designed around a gap in existing apps: poor coverage of Nepali and South Asian dishes. It combines a multi-model AI pipeline with an offline-first architecture so users can log food the way they actually eat — a photo of dal bhat, a quick "two rotis and chicken curry," a voice note, or a barcode — and get instant calorie, macro, and micronutrient estimates.
Every action is written to a local database first and synced to the cloud in the background, so the app stays fully responsive with no connection and reconciles automatically when back online. On top of the core tracker, CaloGlow includes an AI nutrition coach, personalized calorie/macro targets, water and step tracking (with Google Fit sync), trends and streaks, a custom-recipe library, and a subscription tier — all wrapped in a polished, themeable design system.
Key features
Multi-modal AI meal logging — photo (vision), free text, voice, and barcode, with a layered model fallback chain for reliability and a local dictionary that includes Nepali dishes.
Offline-first sync — local-first writes to IndexedDB with background synchronization, conflict handling, and tombstone-based deletes so nothing is lost or resurrected.
Micronutrient breakdown — AI-estimated vitamins and minerals against daily reference intakes, beyond just calories and macros.
AI nutrition coach — a conversational assistant for guidance and questions.
Personalized targets — TDEE and calorie/macro goals computed from user body stats and goals (Mifflin–St Jeor).
Health integrations — water logging, step tracking, Google Fit sync, weight tracking, and optional menstrual-cycle tracking.
Engagement & growth — trial engagement scoring with behavioral nudges, referrals, and a shareable "send a meal" feature.
Subscriptions — free trial plus monthly/annual plans via a Merchant-of-Record provider, with feature gating.
Tech stack
Frontend: React, TypeScript, Vite, Tailwind CSS (token-based design system with light/dark theming)
State & storage: Zustand (persisted), Dexie / IndexedDB (offline-first local database)
PWA: vite-plugin-pwa with Workbox service worker (network-first API caching, auto-update, installable)
Backend: Supabase — authentication (incl. anonymous/guest sessions and PKCE), Postgres with Row-Level Security, and edge functions
AI: Multi-model pipeline (Gemini 2.0 Flash, Groq Llama, Claude) for text parsing, image/vision analysis, and the coach
Integrations: Open Food Facts API, Google Fit (PKCE OAuth), Paddle (payments / subscriptions), Resend (transactional email)
Highlights worth calling out
Designed and shipped a resilient offline-first sync layer (local-first writes, background reconciliation, tombstones) rather than a naive online-only CRUD app.
Built a fault-tolerant AI pipeline that degrades gracefully across multiple providers and a local fallback, so meal logging never hard-fails.
Solved a real localization gap by tailoring food recognition and a seeded catalogue to Nepali / South Asian cuisine.
Implemented an end-to-end subscription and growth system — trials, engagement scoring, referrals, and sharing — not just the core tracker.