Hi everyone. I've started building this while working on my small Steam game, 'cause was trying to choose tags for myself.

Eventually it became a separate tool. The analyzer runs in the browser using SQLite/WASM, small embedding models exported to ONNX, and a LightGBM ranker exported to JSON. Training and data collection happen offline; the site serves static files.

The article covers the implementation and the mistakes along the way.

Dashboard alongside with all parsed RAW Steam (SQLite snapshot) are on Github: https://github.com/Suvitruf/steam-tags-analyzer