The problem
Chatbots answer with a wall of confident text, and you can't tell what is verified and what is invented. Surfyplex turns a photo of a coat, a watch or a plate into a map you can walk through, one branch at a time, with every claim labelled.
What we built
- Start from a camera photo, an uploaded image, text or voice
- AI photo analysis that identifies the object and proposes directions to explore
- An interactive map in 2D, with an optional 3D "knowledge universe" view
- "Go deeper" cards streamed in parts and labelled fact, analogy or hypothesis
- Corrections ("it's not a brooch, it's...") that rebuild the map, and follow-up evidence photos
- An "Objects" mode that reads maker's marks and explains how to describe an item for resale
- Saved library synced across devices, shareable maps with link previews, installable app in Russian, English and French
How it's built
Each exploration and the spending ledger live in their own small database objects on Cloudflare, so AI costs can never run past a hard limit. Several AI models were compared blind and the one with better answers was kept, not the cheapest.
The whole funnel is instrumented step by step, and the first paid acquisition test ran in the first week.