This chess variant's theme is evolution. One morning in early 2016 I woke up with an idea: what if chess were built around promotion? Eventually, this resulted in the following game:
It plays like chess, but you start with only pawns and a king. Pawns can be promoted to bishops or knights, while those minor pieces can be promoted to rooks. Their promotion rights accumulate (every three pawn moves, pawn promotion right; every three minor piece moves, minor piece to rook right). Rooks are transformed to a knight or a bishop on the rook's fifth move.
Recently, I built a playable version. The AI opponent works on phone with no internet. The search engine uses a lightweight neural net. The engine was first weak and my friends complained about it.
So I have spent quite some time to train a better NN for the board evaluation. Fixed some bugs in search. And I have even designed a "Fun" mode where the AI opponent is thinking while you are thinking. (WARNING: "Fun" mode is ruthlessly draining the battery)
Now, when the engine became stronger, my friends complain it is hard to beat. Is the Fun mode less fun now? Should I allow the engine to lose?
More in the writeup: https://penkovsky.com/post/evochess/
Code: https://github.com/penkovsky/evochess/