# Ibo

- **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:** [junho kim](https://cerebralvalley.ai/u/gnar)
- **GitHub:** https://github.com/kjhh2605/Meta-Hackathon
- **Demo video:** https://youtu.be/4NLaaIOFXkc
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4170

Ibo is an Android app for street-market vintage shopping: find a piece from your photos, try it on with clothes you already own, and walk back to the stall where you saw it.

I shop for vintage clothes a lot, and the app uses Dongmyo, a Seoul street lined with vintage shops and stalls, as its example setting. Three things keep getting in the way there:
1. Where was it? Stalls have no signs or addresses. An hour later you can't tell which stall had the jacket, or which of your forty photos shows it.
2. Will it go with what I own? There are no fitting rooms and no product photos to try on virtually.
3. What about with that other piece? Every item is one of a kind. Once you walk away to compare, it may be gone.

How Ibo works:
- Shopping mode: start a trip, snap photos, and mark "New shop" at each stall. Each photo stores its location and is indexed on the phone; Muse Spark describes each storefront.
- Find it: type or say "the blue jacket in the second shop". A Muse Spark agent plans the search with tool calls over your trips, shops and local photo matches, and SAM 3.1 outlines matching garments. The top 3 candidates come first, then "Show all".
- Pick: tap a candidate, or tap the garment yourself (on-device EdgeTAM). It becomes a saved cutout that remembers where you photographed it.
- Try on: fill outer/top/bottom slots from saved pieces, your wardrobe or trip photos and tap Fit. Your face and hair are masked on the phone first, Muse Image renders the whole outfit in one image, and your face is restored on the phone.
- Walk back: a walking route to the spot where you took the photo. The photo is the address.

Privacy: the original self photo, raw location trail, photo embeddings, wardrobe originals and face-restored results never leave the phone. A relay holds the API keys and does not store image, audio or coordinate payloads. If masking fails, the upload is blocked.

Built Oct 26 - Nov 4, 2026: the app, the relay, native plugins, tooling and the demo video. Pre-existing: the Astryx design system, a Capacitor app template, and Meta's open EdgeTAM and PE-Core weights. In the demo video, app screens come from the web mock build with sample data (labelled on screen); the "Your look" panel is a real Muse Image result from a live run on the Android emulator.

## More from Meta Global AI Developer Hackathon

- [Juro](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4167)
- [Multi-turn](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4168)
- [LifeLens — Your Life, Answered](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4169)
- [Eden](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4171)
- [ClearHire](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4172)
- [KRAI-A](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4173)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4170. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt