# seamproof

- **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:** [Hary Merdeka](https://cerebralvalley.ai/u/hary178)
- **GitHub:** https://github.com/harymerdeka/SeamProof
- **Demo video:** https://youtu.be/YH62FWsAf1o
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1624

SeamProof is an agentic workflow that takes a small garment workshop from design to pattern to cut, end to end.

The problem: customers rarely send a pattern. They send 2–3 photos ("make this, but with the back of that one") or just describe what they want. A small workshop guesses the pattern, sews a sample, finds the mistake and cuts again. One wrong guess wastes fabric, hours of sewing and days. Indonesia alone has millions of people working in textiles and garments, mostly in small shops.

What SeamProof does:
1. DESIGN: start from a written prompt (Muse Spark returns two design options) or from photos (Muse Wonder reads each photo into parts; SeamProof merges collar and pocket from one shirt with the back yoke and pleat of another). Refine one change at a time in plain English; every change is versioned. A 3D preview and flat sketch are drawn by code from the same categories as the pattern, so the preview never shows a detail the pattern cannot make.
2. PATTERN: Python drafts every line, then 15 deterministic checks run. If a check fails, the agent loop starts: Muse Spark proposes a parameter change → range check → edit-locality check → redraft → verify, until it passes (max N rounds, every round logged). Muse Spark Open explains the result in tailor language. From a photo of the first sewn sample (toile), Muse Wonder spots fit problems and a rule table sizes the correction into a verified version 2. A pattern maker signs off before the cutting file unlocks.
3. CUT: production pack with quantities per size, fabric metres, cost, cut list and a marker across the fabric width, ready to print 1:1, plus Muse Spark notes for the cutting room.

Rule of the project: models read, propose and explain; code draws and proves; a person signs. Before the hackathon only the workshop , instagram @pintar.jahit and one sample shirt existed; all code was written during the hackathon

## More from Meta Global AI Developer Hackathon

- [Antonio Fernández García](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1621)
- [PebsTeam](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1622)
- [Night Sky Finder](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1623)
- [Agrisense](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1625)
- [dispatchly](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1626)
- [Nodesnappers](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1627)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1624. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt