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.
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