Extract tactics automatically

[Fulvio]

Both lichess and chess.com tactics training is repetitive and boring.
What I want:

• 1. Extract from a database the position where the best move is significantly better than the second best
• 2. Use unsupervised machine learning to assign the position to a category
• 3. Pick the next tactics for the categories that I don't have already easily solved

The first point I can easily achive with a scid script. But processing a database would take time, so I decided to use the lichess's puzzle (https://database.lichess.org/#puzzles ) for testing.
The second point is the hard part:

• do you know if there is already some model available?
• I asked chatgpt to create the code for a clustering algorithm ( https://github.com/benini/extract_tacti ... stering.py ). But it is slow and not really accurate: do someone have expertise in those type of algorithms and can give any suggestions?