# Ive

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Nurdaulet Yerk](https://cerebralvalley.ai/u/Nurdaulet)
- **GitHub:** https://github.com/NurdauletYN/ive-meta-hackathon
- **Demo video:** https://youtu.be/WuehTiWzUfk
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5110

Rocket is a harness that turns an app idea into a store-checked Flutter app. The harness is the rocket; the model is the pilot.

Problem: after 15 published apps I kept rebuilding the same plumbing: sign-in, onboarding, a paywall with subscription disclosure, restore purchases, privacy and terms screens, account deletion. Each rebuild brought the same bugs and store-review problems. An AI builder that regenerates all of it every time repeats those bugs.

What I built this week: a locked Flutter core the model cannot edit (auth, onboarding, paywall, settings, legal, account deletion) plus an editable surface. Muse Spark returns a schema-validated app spec and catalog; a deterministic writer and a design kit (11 palettes with WCAG AA contrast, bundled fonts, product photos) turn it into screens, so output is styled rather than stock Material text lists. Every file operation passes a guard. A gate then runs flutter analyze, tests, a web build and 15 store-readiness checks, with up to two repair rounds fed only by gate failures. A local studio chat builds an app, then refines it, with projects and versions. A CLI prepares the store listing, a privacy template and the icon, and can build and upload an IPA to TestFlight using the user's own Apple key, which never reaches a model, a log or the browser.

Pre-existing vs new: Ive, my private AI Flutter builder, is not in this repo. Everything here was written during the hackathon.

Limits: auth and billing run in demo mode unless the user supplies Firebase or RevenueCat config; the TestFlight upload is tested against a fake runner only; the checks are engineering gates, not an App Review prediction; a harness-versus-no-harness comparison is implemented but not run at scale, so I make no claim about it.

## More from Meta Global AI Developer Hackathon

- [AURA RLVE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5107)
- [Slate HQ Ltd](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5108)
- [Rabin Singh](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5109)
- [Mantis Analytics](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5111)
- [BSDEV](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5112)
- [Not Random](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5113)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5110. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt