alpkeskin/locus ? reverse-engineered prompt

Build me a web app called Locus that can take a street photo and guess where it was taken, then show me the GPS coordinates and a confidence score.

I want a simple frontend where I can upload an image, start a search, and see the result on a map with a link to open it in Google Maps. I also want a training flow where I can enter a center point, radius, and number of sample points, then build a location database from street view images.

Use the same general approach from the README, detect buildings in photos, extract visual features, and compare them against a stored vector database so it can match similar places. Set it up so the whole thing runs locally with Docker, with a backend API and a clean frontend. Add basic docs for setup, environment variables, and how to use the train and search endpoints. If you need current API details or model docs, look them up online.