# Inspec

- **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:** [Felix Deng](https://cerebralvalley.ai/u/mallard64)
- **GitHub:** https://github.com/Mallard64/inspec
- **Demo video:** https://youtu.be/kR0VSdftN_c
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2503

When a box of parts or food is packed, someone has to check it matches the packing list. People get tired and miss things, and most automated checkers need weeks of training for each product, which small-batch factories can’t afford.

Inspec skips the training. You type the packing list, and it checks a photo of the box against it. One AI model (SAM 3.1) labels every item in the photo. Plain code then counts and measures them, so it can say exactly why a box failed, like “zero long screws, two short.” Another AI model (Muse Spark) turns the list into checks, fixes any check that wrongly rejects boxes it knows are good, and judges damage, like a torn peel, which can’t be measured. Boxes it’s unsure about go to a person.

On the MVTec LOCO AD dataset, Inspec caught 215 of 220 boxes with wrong parts, sizes or amounts. Muse Spark alone caught 198. On damage the two perform about the same, so I’m not claiming Inspec is better there.

## More from Meta Global AI Developer Hackathon

- [Bamalin](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2500)
- [ostefani](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2501)
- [SkedCast](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2502)
- [Borderless](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2504)
- [StyleCatcher](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2505)
- [Grey-matter47](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2506)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2503. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt