Daniel Keck
Weather apps make sense out of millions of parameters, but they cannot tell you what the person standing one valley over is actually looking at.
Core trick
Projects each photo's compass heading into a view-cone on a map, then checks the image against radar and satellite data for verification.
Description
Weather apps make sense out of millions of parameters, but they cannot tell you what the person standing one valley over is actually looking at. The forecast in your pocket is the forecast for everyone in your ten-kilometer cell.
VYU turns every phone into a directional weather webcam. A photo plus the compass heading from the device's orientation sensors becomes a view-cone on a shared map, pointing exactly where the camera was aimed.
Three Claude Managed Agents on Opus 4.7 do the work, on two clocks. When you submit a photo, a Classifier reads it against a phenomenon taxonomy and a Reconciliation agent fires alongside, checking the image against DWD radar, EUMETSAT MTG infrared and lightning, and the Open-Meteo forecast. What comes out is a verified report. When you ask a question, a Deep Researcher answers it against your own report and a per-cell memory of what other contributors nearby saw. If the answer mentions the cool wind off the lake, it is there because someone there felt it. The agent inherits.
One photo earns one question. A fair trade. Tit for tat.
README excerpt
# VYU > **Retired 2026-10-23.** The site at vyou.dkck.dev is offline and its backend (Supabase project, Claude Managed Agents, Vercel project) has been deleted. This repository stays, read-only, as the record of the submission to the [Built with Opus 4.7 Claude Code hackathon](https://cerebralvalley.ai/e/built-with-4-7-hackathon) (Anthropic × Cerebral Valley, 21–26 April 2026). The code describes a system that ran during that week; nothing in it is live today. *What's in your view.* VYU is a community-fed situational weather map. A field observer takes a photo of what they see — a storm, a cloud structure, a clean front edge, a sky over a Bavarian monastery — and VYU captures not just the image but the compass heading, so every report becomes a **view-cone** on a shared map pointing exactly where the camera was aimed. Three Claude Managed Agents on Opus 4.7 do the work behind the cone — but on two clocks: two when you submit (the Classifier reads the photo, Reconciliation compares it against DWD radar + EUMETSAT MTG satellite + Open-Meteo), and the third when you ask (Deep Researcher answers a place-grounded question against the verified report and a per-cell memory of prior contributors). The product lives in one motion: a sky photo becomes a directional, AI-verified weather report on a shared map, and your contribution earns you a question only that map can answer. The name is pronounced *"view."* The **V** doubles as the heading-cone geometry; the **YOU** puts the obse