so my friend made a py chess game and it had an unbeatable ai this wasn't a surprise to me however I have heard of Alpha go and the complexity of the game and I am wondering is it possible to replicate this in repl.it?
No, you won't be able to replicate this on repl.it. According to the white paper, 44 million games were played over 9 hours. The games were played using Google's custom TPU (Tensor Processing Units) manufactured specifically for machine learning. Because your repl (neither your own computer) is not even close to the computational power that Google used for their calculations, you won't be able to train a model, even at a substantial fraction of the skill level.
There is an open source adaptation of Alpha Zero called Leela Chess Zero. It's one of the best chess engines out there and many people have had to offer their computer compute time to get to that point.
However, don't be discouraged! You may still be able to create your own algorithm with a smaller network. You may also consider pretraining your model someplace else, and developing an interface for that model on replit. Or, you can use something like TensorFlowJS to do your machine learning on the client (rather than on replit servers).
Maybe you would want to consider creating a chess ai that does not involve neural networks
Hope I could help =P