Project · 2025
GeoHelper
Desktop GUI that takes a screenshot, runs it through an LLM-vision analyser, and tries to figure out where the image was taken.
A desktop tool for geographic-clue extraction. Capture a screenshot, feed it through an LLM-vision pipeline, and get back the model’s best guesses about location, landmarks, vegetation, road markings, and whatever else is visible.
The codebase is split by responsibility — capture/ for screen
grabbing, ai/ for LLM analysis, gui/ for the desktop interface,
scrapers/ for any reference data the analyser pulls on, and
data/ for what gets stored locally. A __main__.py ties it all
together so it runs as a single launchable app.
Mostly a “see how far you can push vision models on real-world context clues” experiment.