Side-project: See-to-Solve!

See-to-solve is a tool to identify the right move for turn-based games, starting with Chess. The original idea was to use machine learning-trained models to:

1. Identify the game
2. Identify if the “state/position” has changed
3. Identify the position in the game (including who’s turn it is)
4. Recommend the best move (or moves)

I was planning to run this on my Ray-Ban Stories glasses.

My motivations for this project were:

• a. See if I could do it
• b. Learn how to use PyTorch
• c. Reduce my desire to play online chess!

The current version leverages many open-source projects to build steps #3 and #4 for online chess (source). The three main components are

• i. a chrome-extension to save an image of a board from online chess websites
• ii. a PyTorch model to identify the chess position in FEN
• iii. a python version of the Stockfish chess engine to recommend the best moves