# ALBA

- **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:** [mehmet salih altin](https://cerebralvalley.ai/u/msaltin)
- **GitHub:** https://github.com/mehmetsalihaltin-max/sez-dogrulama
- **Demo video:** https://www.youtube.com/watch?v=q8_vUpeXwL0
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1730

SEZ is an evidence gate for airline operations centres. A duty manager asks a plain question, typed or spoken, such as "How is the Singapore to Bali route for 20 June 2025?". SEZ answers by showing how strong the evidence is for each event that could affect the route. The team makes the call; SEZ feeds that decision with graded, quoted evidence.

The problem is dirty information. News, notices to pilots (NOTAMs) and satellite images are scattered, and a wrong report costs as much as a missed one. In SEZ every event becomes a checkable claim. Meta's Muse Spark searches the news in local languages and must quote the exact words that support or contradict the claim. Our code then checks that the quote is really there, that the date is inside the window and that the place is named, and sets the evidence level itself. "Confirmed" means supported by at least three different news publishers, counted by distinct domain after removing aggregators and re-posted copies. We call these independent publishers; code cannot prove editorial independence. SEZ shows news confirmation and official confirmation separately and suggests which official body to check.

NOTAMs are read live in code. Satellite passes add a second channel, and in a flooded-runway scene (Porto Alegre, May 2024) SAM 3.1 outlines the runway so water on it can be measured.

We tested SEZ on 15 claims: 7 real October 2026 events and 8 altered ones. It rejected all 8 altered claims with zero false confirmations, and supported 4 of the 7 real ones. The full run output is in the repository.

## More from Meta Global AI Developer Hackathon

- [The Ground](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1727)
- [Lewis Muguna](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1728)
- [SEO Noobs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1729)
- [VsionX](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1731)
- [MUSE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1732)
- [SYSVIA ProofSight](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1733)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1730. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt