NNUE is the interesting part to me. Alpha-beta tree search is useless without a good value function. Not sure what would be the best way to generate the training data if you're starting from scratch.