# Fight Club

- **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:** [veerabhadra yerram](https://cerebralvalley.ai/u/veer7), [Shiva Sai](https://cerebralvalley.ai/u/tonystark69), [Sai Koushik](https://cerebralvalley.ai/u/LouisBloom)
- **GitHub:** https://github.com/saikoushikpallapolu/Freshtrace_teamfightclub_meta
- **Demo video:** https://youtu.be/9Nzze3zq7v8
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2596

A crate-level warning should lead to an item-level investigation—not guesswork. Our project is a produce inspection and stock-priority assistant that helps warehouse operators connect sensor warnings with visual evidence, investigate potential spoilage, and document their decisions.

The intended workflow begins when an elevated volatile organic compound (VOC) signal flags a crate for inspection. The operator opens the crate, photographs its contents, and saves the image against the batch. Our implemented visual pipeline then uses Meta SAM to segment individual tomatoes and Muse Spark to describe their visible condition. Operators can select a region, examine its contextual crop, understand why it warrants attention, and record a separate human review.

The product connects three complementary experiences:
- Rescue: prioritize batches and items that warrant inspection.
- Investigator: guide follow-up checks when evidence is incomplete or ambiguous.
- ShelfWatch: preserve batch history, including photographs, model findings, and human reviews.
The distinction is traceability: a finding stays connected to the image region and evidence that produced it. Human corrections are recorded separately, preserving the original model output.

Our working prototype supports persistent batch records, validated photo uploads, real SAM/Spark inference, selectable segmentation overlays, durable analysis jobs, and saved reviews. Full sensor-to-batch integration and the complete stock-action workflow remain in development.

The goal is to help operators focus their attention earlier, avoid indiscriminate crate-level decisions, and build a more accountable inspection process. Waste reduction and operational benefits will be evaluated through controlled physical trials.

## More from Meta Global AI Developer Hackathon

- [Infinity](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2593)
- [Pawprint Club](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2594)
- [Jeffery Nketiah](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2595)
- [SignBridge](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2597)
- [Ode: Give Muse a body](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2598)
- [The Gatekeepers](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2599)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2596. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt