# Silvi

- **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:** [Djimo Serodio](https://cerebralvalley.ai/u/metafarmer)
- **GitHub:** https://github.com/SilviProtocol/silvi-survey-agent
- **Demo video:** https://youtu.be/a3-Qj559fDA
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4668

I run Silvi. We pay for verified tree-growing results, and until now the proof has been a geotagged photo. A photo with a GPS point tells you a tree stood somewhere near that point on that day. It cannot tell you which tree it is, and it cannot bring anyone back to that same tree next year to check on it. Anyone who has tried to verify forest work knows that gap well, because money moves on those photos. That's why survey-grade equipment exists, with LiDAR rigs and RTK receivers, and it's out of reach for most of the people growing trees. We're setting out to get that same high-fidelity, faithful record of the evidence from an affordable Android phone or iPhone.

This week we built the thing I have wanted in the field for years. You hold up a phone and walk. The phone works out where it is from the camera, finds each trunk as you pass it and keeps building confidence in that tree from every new angle you give it. When it is sure, a ring closes around the trunk and the tree gets a number. If you want to mark a spot yourself you place a marker by hand and drag it around, and you can talk to the agent while all of this is happening.

The demo is one walk of about 170 seconds in a backyard, on a Pixel with no LiDAR. The phone locked in three plants and saved two markers I placed by hand, all in the same 3D space. At the end I said, "Take me to the first tree that I registered today," and the phone guided me back through the yard to that tree. That walk back is the whole point. The record stopped being a photo and became a place the phone remembers.

The right side of the video is the same walk rebuilt afterwards from its own recorded data, running on the same clock as the phone footage. The point cloud was cleaned offline, and the photos pinned to each tree in that view were cut from the walk's own video.

Before this week we had geotagged photos, our own tree detector and an early positioning app. This week we added the voice agent, the walking survey and the 3D console, and positioning got a lot better.

From here we take it into denser forestry work, and then to the part I care about most, walking back to the same trees year after year and watching the record grow up with them.

## More from Meta Global AI Developer Hackathon

- [SLVRLeaf Digital](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4665)
- [SN2.0](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4666)
- [Fireshift AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4667)
- [CargoMesh DockMind](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4669)
- [khanaitechnologies](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4670)
- [jackpot-joe](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4671)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4668. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt