I built a free AI nutrition tracker as a solo dev — here's what I learned

A solo developer built goNutriTrack, a free multilingual PWA nutrition tracker that uses AI for meal logging via photo, voice, barcode scan, and text search. The app, powered by Anthropic Claude and a serverless stack, overcame challenges like iOS PWA white screens and offline data queuing. The developer emphasizes shipping early and notes that PWA on iOS remains painful.

What I built goNutriTrack is a free multilingual PWA nutrition tracker. The idea was simple: make meal logging so fast that people actually stick with it. You can log meals by: 📸 Photo — Claude Vision identifies ingredients automatically 🎤 Voice — say "200g chicken and rice", AI parses it 📷 Barcode scan — OpenFoodFacts database 🔍 Text search — OpenFoodFacts + USDA Plus: AI Coach, workout tracking, weight tracking, water tracking, PDF export. Works in 5 languages Greek, English, German, French, Spanish . The stack Frontend: React + Vite PWA Backend: Cloudflare Workers proxy for Anthropic API Database: Supabase AI: Anthropic Claude Vision + Haiku for voice parsing Hosting: Cloudflare Pages No server, no Docker, no DevOps. Everything serverless. Technical challenges - iOS PWA white screen The hardest bug. When users add the app to Home Screen on iOS, sometimes it opens to a blank white page. The fix was switching the Service Worker navigation handler from network-first to cache-first stale-while-revalidate , so iOS always has something to show immediately. - Offline queue Supabase calls fail when offline. I built a localStorage queue that stores pending inserts/deletes and syncs automatically when the connection returns — without the user noticing anything. - Vite + PWA build pipeline The Service Worker cache name needs to change on every deploy to force updates. I wrote a Vite plugin that patches the SW file during config not buildStart to avoid race conditions with vite-plugin-pwa. - AI meal recognition accuracy Claude Vision is good but not perfect. The solution was making every recognized item editable before saving — users can adjust portions and correct names. This turned a potential frustration into a feature. What I learned Ship early. I spent weeks polishing before showing anyone. The first real users found issues in 10 minutes that I had missed for months. PWA on iOS is still painful. Safe area insets, white screens, push notifications not working — Apple makes this hard on purpose. Solo means every decision is yours. That's both the best and worst part. What's next Looking for honest feedback — especially on the AI logging accuracy and anything that feels broken. Try it free: gonutritrack.com