# Eucher Abatti

- **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:** [Eucher Abatti](https://cerebralvalley.ai/u/Tobi229)
- **GitHub:** https://github.com/T0b0i7/NaviGarde_IA
- **Demo video:** https://youtu.be/sT0u5a6nzxw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2590

NaviGarde IA is a maritime safety platform built for Beninese waters, where boat inspections are still largely manual, paper-based and untraceable. The problem is concrete: an overloaded canoe departs without anyone measuring its actual capacity, and when an incident happens, no decision can be reconstructed.

The platform covers four points in the operational chain. Inspection: an agent photographs the boat, the model analyses the images and proposes a passenger count, lifejacket visibility, hull and engine condition, a risk score and an explicit confidence level. Alerts: every anomaly confirmed by a supervisor automatically generates an urgent alert, and a public page lets a sailor without an account report distress. Reports: generated as archived PDFs with a SHA-256 fingerprint and versioning. Audit: every sensitive action is recorded with its author, address and timestamp.

The design principle is that the AI proposes and the human decides. No detected anomaly becomes a decision without named human review: the supervisor confirms, corrects or dismisses the analysis. Departure is forbidden as soon as the risk score crosses the critical threshold. The system never presents a hypothesis as fact: anything not visible in a photo is marked as such and listed as an uncertainty.

The project targets field agents, supervisors and administrators of maritime authorities, with four roles whose permissions are enforced server-side, not merely hidden in the interface.

Built with React, Express, tRPC and MySQL with Drizzle ORM, the project has 20 tables, 10 versioned migrations, 62 automated tests and a browser sweep checking all four roles. Meta's model is called exclusively server-side, for four uses: inspection image analysis, report writing, voice alert triage and explanatory reading of dashboard metrics.

## More from Meta Global AI Developer Hackathon

- [Skillow](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2587)
- [SpaceGrade](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2588)
- [Swl Servoces](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2589)
- [Resonance Labs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2591)
- [StormGate-Zero](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2592)
- [Infinity](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2593)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2590. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt