VITAL

Built in 28 days for the Replit Agent 4 Buildathon

VITAL is a full mobile fitness super-app — workout logging, AI food scanning, GPS run tracking, water and sleep counters, gamification, social challenges. It went from npm init to TestFlight in 28 days, all of it shipped from a Chromebook. No Mac involved.

The features that mattered

• AI food scanner — point the camera at any meal, get protein/carbs/fats/calories with confidence intervals. Uses a multimodal vision model on the device's built-in camera, with a small fine-tuned head for portion estimation. Caches recognized meals so the second time you eat the same lunch it's instant.
• GPS run tracker — full route capture with replay, splits, elevation, pace. Works offline, syncs when back online. Battery cost is <6% per hour of tracking.
• Custom muscle-map SVG — interactive workout planner where you tap muscle groups to highlight exercises. Built as a single hand-drawn SVG with per-muscle hit areas. Animates fill on selection. Twenty hours of work for a feature that became the visual signature of the app.
• Ramadan + halal mode — fasting hours auto-calculated by location, calorie targets adjust dynamically, suhoor/iftar reminders. Halal meal database included by default. The fitness app market routinely ignores Muslim users; VITAL doesn't.
• Streaks, levels, achievements — restrained gamification (no infinite-engagement traps). Streaks reset only on full inactivity, with one "freeze" per week.

The Mac-free build pipeline

The whole app was built on a $300 Chromebook. The trick: Expo's EAS Build service runs the actual macOS compile on Expo's servers. npx eas build --platform ios --profile production queues a build, the cert/provisioning-profile dance is handled by EAS, and ~15 minutes later you have a signed .ipa ready for TestFlight. eas submit uploads it. I never opened Xcode.

For development I used a real iPhone with Expo Go, hot-reloading from the Chromebook over Wi-Fi. Faster than any simulator. Full write-up here.

What I'd change in v2

The current AI food scanner uses a hosted vision API. For v2 I'd move to on-device inference with Core ML / TensorFlow Lite. Privacy is the killer feature for a fitness app — users genuinely don't want their meals leaving the phone, even encrypted. The model size is fine for modern phones; battery cost is negligible.

The deliverables shipped

Beyond the app itself, VITAL came with a marketing landing page, full Figma design system, a 14-slide pitch deck, App Store screenshots in multiple device sizes, and a Loom walkthrough of the architecture. The Buildathon judges were impressed; more importantly, the package is what a founder actually needs to validate a fitness app idea, not just code.

If you're thinking about a mobile app

Most JS devs avoid mobile because the toolchain looks scary. Expo + EAS Build collapses that toolchain to "ten npm scripts and a paid Apple Developer account." If you have an idea that needs to be on iOS, you're a week of focused work away. More on how I scope mobile builds.