# PackMatch

- **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:** [Alonso Cespedes](https://cerebralvalley.ai/u/alonsocespedes)
- **GitHub:** https://github.com/alonsinhoiu/packmatch
- **Demo video:** https://drive.google.com/file/d/1pwzU9oI94BH9_eEaXDst3LyasQlApgRk/view
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1092

PackMatch helps importers and logistics operators catch discrepancies between commercial invoices and packing lists before dispatch, with visual evidence and an actionable supplier message.

Live demo: https://packmatch.netlify.app

A shipment can look correct at the document level while hiding a short delivery or an incorrect line amount. PackMatch brings both documents into one review workflow: diagnosis, side-by-side source documents, a product matrix, and a bilingual supplier memo with CSV export. Findings link back to their original invoice lines and packing rows through highlighted bounding boxes.

The demo's C02 case shows SKU-401 billed at 500 PCS while three packing rows sum to 480 PCS (200 + 200 + 80), a shortage of 20. SKU-402 has an arithmetic error: 100 x USD 12.50 should total USD 1,250, but the invoice declares USD 1,350. C06 demonstrates human review: SKU-806 has 90 SETS without a printed conversion factor. The operator explicitly confirms 1 SET = 5 PCS, and the action is retained in the audit log.

Meta Muse Spark 1.3 reads the documents; deterministic TypeScript rules reconcile quantities, amounts, weights and packaging; people resolve ambiguity. The interface supports Spanish, English and Simplified Chinese, independently from the English/Spanish memo language. Spreadsheet and manual input use the same workflow.

The public repository includes seven synthetic cases and a reproducible comparison expecting 13 findings with exact source references. This is a small regression dataset, not a general accuracy guarantee. Bundled examples run without an API key; live extraction requires server-side Meta API credentials.

Built by Alonso González Céspedes in Chile for Build for Your World.

## More from Meta Global AI Developer Hackathon

- [CRYPTOR.AGENCY](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1089)
- [First Pass](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1090)
- [EdAI Systems](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1091)
- [AnaliZac](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1093)
- [Koanic](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1094)
- [Musavir](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1095)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1092. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt