steamvectors

Vectorized embeddings, visualization, and recommendations of Steam games

Pipeline

• Download data from 🤗 -> raw_games.parquet
• Clean data, removing html & md, converting lists to strings, removing duplicates -> games.parquet
• Embed games.parquet using OpenAI text-embedding-3-small (512 dimensions) -> embeddings.npy
• Upload embeddings -> NeonDB + pgvector

Backend

Python + FastAPI

• Player taste embedding generated using top played games weighted by playtime and normalized
• Queries pgvector db

Frontend

React + Vite + Tanstack

• Allows users to login with Steam OAuth
• Visualization of vectorized data with react-three-fiber
• Shows recommended games, pathways to such, and highlights most played/searched for games

Hosting

TBD