# Fobby

- **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:** [Eric Hsu](https://cerebralvalley.ai/u/erichhsu)
- **GitHub:** https://github.com/fobby-io/fobby-connect
- **Demo video:** https://youtu.be/05ZM_NucMVU
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3097

Fobby Connect reads freight documents (packing lists and shipping orders) from any sender into one fixed schema, and learns each sender so the next document needs less review.

Forwarders retype these PDFs by hand today, 30–60 minutes a document, because every sender prints its own layout: carton ranges instead of counts, different labels for the same field, ROC-year dates, KGS or LBS. Plain OCR returns text, not checked values in the right fields.

Muse Spark reads each page into a strict JSON schema and copies only what is printed. Code checks row sums against the totals, pallets, PO numbers and container check digits, and a page that fails is read again with the findings as a hint. A second system checks every value against the page. A document from a known sender that passes every check is approved and sent on with no person involved; anything else goes to a clerk, who reviews it beside the PDF. When the clerk corrects a misread, Muse Wonder rewrites that sender's prompt, and the new prompt goes live only if replaying the sender's approved documents shows no regression. Approved data goes out as CSV or JSON, through a signed webhook, or by email.

Live: https://connect.fobby.io (three synthetic sample documents are already in the list). Open source under AGPL-3.0.

## More from Meta Global AI Developer Hackathon

- [Storieta](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3094)
- [Team CaféPilot](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3095)
- [Nathan Romano](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3096)
- [AsinZen](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3098)
- [PARALLEL](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3099)
- [aipool](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3100)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3097. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt