# LAVERTY – Infinity AI Technology Labs

- **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:** [Mohamed Maamoun](https://cerebralvalley.ai/u/mohamedmaamoun)
- **GitHub:** https://github.com/infinity-ai-labs/laverty-case-reader
- **Demo video:** https://youtu.be/QuHnBGJvUgk
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1508

LAVERTY Case Reader turns a scanned Arabic court file into a bilingual case summary a lawyer can trust.

The problem: in the UAE, court papers reach lawyers as scanned Arabic documents covered in stamps, signatures and handwritten notes. A next hearing date may exist only in a clerk's handwriting, a printed amount may be crossed out and corrected by hand, and a claim may cite an article the law doesn't have. A lawyer must read every page before any legal work starts, and one missed note can cost a deadline.

What we built during this hackathon (a new feature for LAVERTY, our Arabic-English legal AI for UAE lawyers):
1. SAM 3.1 finds every stamp and signature on every page.
2. Muse Spark 1.3 reads all pages together, plus a close-up of every mark, and returns a fixed structure in Arabic and English: court, parties, dates, deadlines, claims, key facts and citations, each linked to its page. A fact known only from handwriting or a stamp says so; a handwritten correction is shown with both values.
3. Handwriting gets a second look: Muse Spark finds the notes SAM 3.1 misses and places each one again from a close-up.
4. Every cited article is checked against the official UAE law text: right law, article exists, quoted wording matches word for word. Anything it can't verify is labelled "not checked", never passed.
5. The summary opens with "Check these first"; on the demo page, clicking any date, citation or mark opens the page it came from and highlights it.

How we tested it: 13 fictional scanned case files (33 pages) with a hand-checked answer key and traps planted on purpose. Real run on Meta's models (run 6): 6/6 citations of non-existent articles caught, the changed quote caught, 80/80 citation verdicts right, 35/35 claim amounts, 65/67 parties, 125/131 dates, 98/109 stamps, signatures and handwritten notes in the right place. Full scores and every miss are in the repository.

Demo page: https://infinity-ai-labs.github.io/laverty-case-reader/

## More from Meta Global AI Developer Hackathon

- [PushtiPlate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1505)
- [Aurrium](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1506)
- [Kawesi Hakim](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1507)
- [Byte Busters](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1509)
- [AskYourData](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1510)
- [uty](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1511)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1508. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt
