The project walks through the full pipeline: manually collecting data from book scans, comparing Autoencoders vs. SimCLR for representation learning, and deploying with ONNX Runtime and Pyodide to avoid server costs.
The included demo recognizes board diagrams and generates Lichess links for analysis.