# Bosque Vivo Chile — ProGarLabs

- **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:** [Giuseppe Proboste](https://cerebralvalley.ai/u/gp-progarlabs), [Jose Schwerter](https://cerebralvalley.ai/u/fg-progarlabs)
- **GitHub:** https://github.com/GiuseppePrbst/meta-global-ai-dev-hackathon-progarlabs
- **Demo video:** https://youtu.be/sVu7jf1nGQI
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4690

Bosque Vivo Chile is an AI-powered field companion that makes Chile's national parks and biodiversity easier to explore through grounded conversations, visual species suggestions and verifiable environmental data. Its promise: explore nature with AI that knows what it knows, and what it cannot verify.

The problem: reliable information about Chile's parks and species is spread across official park pages, conservation lists, tourism guides and scientific databases, each with its own format and level of trust. Generic AI assistants tend to fill those gaps with confident guesses.

What we built between 26-10-2026 and 04-11-2026: a versioned Knowledge Atlas covering 46 national parks, 190 trails reviewed against official pages and 20 sourced taxa; Wilo, a guide that takes voice or text and answers with sources; open-set photo identification that can abstain; and a public API on Cloud Run behind Cloudflare with PostgreSQL/PostGIS, a spending ledger, consent, territory and rate controls, and a hard expiry date. Anyone can try it without an account: https://api-bosquevivo.progarlabs.com/docs (a technical API surface, not a consumer app). The demo video uses an experimental test client against that live API.

Evidence, with denominators: 79 real recordings with WER about 0.10 (one speaker); 16 of 18 end-to-end voice scenarios semantically correct; wrongly accepted negatives fell from 25/37 to 3/37 with verification on 56 fixed cases with weak labels.

Honest limits: identification is a suggestion, not a confirmed result, and has no specialist review. A photo labelled Nothofagus dombeyi was accepted as N. betuloides on the public route. Trail openings are not verified in real time, and there is no licensed trail geometry. The official mobile identify screen is still a prototype using demonstration data. We have no users or partners yet, and we are not affiliated with CONAF. Judge quickstart: docs/submission/JUDGE_QUICKSTART.md in the repository.

## More from Meta Global AI Developer Hackathon

- [KIN Access](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4687)
- [Lakers](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4688)
- [Lecture Studio](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4689)
- [LUXISOFT](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4691)
- [Tuck](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4692)
- [infinATSolutions](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4693)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4690. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt