# Rangrez Labs

- **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:** [Tushar Agarwal](https://cerebralvalley.ai/u/TechTush)
- **GitHub:** https://github.com/TusharTechs/rangrez
- **Demo video:** https://youtu.be/g4Z30UhPtjA
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/833

Every colour you wear was matched by someone, by eye. Before a garment order is dyed, the dye house dyes a small sample, a "lab dip", and couriers it to the buyer. Rejected? Correct and try again: 2 to 4 rounds of several days each. Big mills use spectrophotometers and colour-matching software; thousands of small dye houses in India still work by eye.

Rangrez (रंगरेज़, "the dyer") is a self-driving dye lab that needs only a phone and a printed colour card. It picks a recipe, a cotton strip is dyed (by a $100 ESP32 rig, or by a person with a dropper), a phone photographs it, SAM 3.1 turns the photo into a calibrated colour measurement, and Muse Wonder corrects the recipe until the cloth matches the buyer's tolerance. It reads the buyer's WhatsApp order, produces a buyer-ready lab-dip report with a production plan, previews the garment in the shade, and drafts the reply in the buyer's language.

Results (simulated chemistry, real-looking photos, live Meta models): 0.66 ΔE00 median measurement error from a phone photo with the card; 24/24 shades matched in a median of 2 dips; Muse Wonder matched 8/8 live campaigns where a fixed policy with the same tools matched 5/8; lab memory cut first-dip error from 3.81 to 1.40. The model never drives a pump: a deterministic safety gate, repeated in the ESP32 firmware, sits behind an e-stop. Honest status: not yet run with real dye on real cloth; that is the next step.

## More from Meta Global AI Developer Hackathon

- [Doby's Universe](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/830)
- [ArthNiti](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/831)
- [MuseDeck](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/832)
- [LLM Hub](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/834)
- [Capybara](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/835)
- [JOBAYER ISLAM POROSH](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/836)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/833. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt