# SoloPaddlerLab

- **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:** [Ogem Eychenne](https://cerebralvalley.ai/u/IceTea42)
- **GitHub:** https://github.com/IceTea42/off-my-shelf
- **Demo video:** https://www.youtube.com/watch?v=ZBj-yZWV4rE
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4340

Off My Shelf turns one photo or a short video of a shelf into ready-to-post resale listings, one per item, each sorted into sell, donate or recycle.

Most homes are full of things nobody uses. What stops people from selling them is not the selling, it is the listing work: a photo, a name, a condition, a fair price and a description for every single item. So things stay on the shelf or end up in the bin.

Muse Spark lists the items in each photo, SAM 3.1 masks each one from its noun phrase, then Muse Spark writes a schema-checked listing per item: what it is, its condition and visible defects, a second-hand price range, a title and description in English, French or Turkish, and a triage decision. Product shots are built from the photo's own pixels. Muse Image only fills parts hidden behind other objects, and a pixel lock, a drift check and a Muse Spark review reject any shot that changed the item, so the buyer gets exactly the item in the picture. Each item also gets a 5-second vertical "shelf to studio" clip and a 360-degree turn of its 3D model.

A shelf agent then acts on the whole pack from a typed or spoken goal: "I'm moving on Friday: sell the electronics fast, give the books away." It prices for a quick sale or the best price, groups cheap items into lots and drafts replies to buyers. The agent proposes, the person decides: every action is checked against floor prices, never-sell items and off-platform payment tricks, picked one by one, and can be undone.

It installs as an Android app (camera, microphone, share sheet in and out to Leboncoin, Vinted, eBay or Geev) or can be used as a webpage, works in three languages, caches and traces every model call, and has a spending cap for the public demo. Built solo, from an empty repository, during the hackathon.

## More from Meta Global AI Developer Hackathon

- [BenchmarkWatcher](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4337)
- [Foresight-AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4338)
- [Real One](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4339)
- [PocketDNA](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4341)
- [Chrishna](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4342)
- [ROF12](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4343)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4340. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt