# Unit'em

- **Event:** [RAISE Summit Hackathon](https://cerebralvalley.ai/e/raise-summit-hackathon)
- **When:** Jul 4 at 9:00 AM – Jul 5 at 7:00 PM (GMT+2)
- **Where:** Paris, France
- **Team:** [Mohammed Jaseel Kunnathodika](https://cerebralvalley.ai/u/jaseel), [Solal PM](https://cerebralvalley.ai/u/Solal), [Jerry Dong](https://cerebralvalley.ai/u/JD101X), [eliott boneu](https://cerebralvalley.ai/u/ILUVJERRY), [Pierre-Louis Manchuelle--Lambre](https://cerebralvalley.ai/u/plml)
- **GitHub:** https://github.com/Thinkr1/Unitem
- **Demo video:** https://youtu.be/8Of98QuJyX8
- **Gallery:** https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/50

How Unit'em actually works (for the script): When code changes on either platform, Unit'em doesn't just diff two files — it runs a five-stage pipeline. First, a Discovery stage scans both the iOS and Android codebases and extracts every design fact: colors, spacing tokens, components, and on-screen copy. Second, a Mapping stage automatically pairs up the matching iOS and Android screens, so it always knows which Android screen corresponds to which iOS screen. Third, when a change is detected, three specialized checks analyze it from different angles — one looks at whether a shared design token changed, one looks at whether a platform-specific style/component changed, and one looks for "drift," like an old hardcoded value that's snuck back in instead of using the shared design system. Fourth, all of that evidence — the change itself, the matching code on the other platform, accessibility/spacing checks, and the team's own design rules — gets handed to a Judging agent, which is the one making the actual Propagate / Hold / Flag decision, not just a diff tool guessing. Fifth, once a human approves a Propagate, a Writing agent (using Cursor) generates the actual Android code fix, and then a separate Review/verification step checks that the generated code is safe and correct — for full-screen rebuilds this includes actually compiling the generated Flutter code before anything is allowed to land — before Unit'em opens the real pull request. So "we send the task to Cursor and it writes the fix" is really the tail end of a pipeline: discover the facts, map the screens, judge the change with evidence, and only then generate and verify the code. That's the part worth explaining, because the judgment and verification are what make this trustworthy enough to hand off to Cursor at all

## More from RAISE Summit Hackathon

- [The Notified Bodies](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/47)
- [Om Solo](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/48)
- [OpsAgents Autopilot](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/49)
- [Blue Brick](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/51)
- [Ghost Walk](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/52)
- [ShipHappens](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/53)

---

Markdown version of https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/50. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt