It's just a repetitive sorting puzzle game. Don't expect any relation to machine learning other than the occasional "This node isn't 100% accurate but it's faster than the one we make you usually use and you need it to beat the par time." I played the game for nearly 3 hours and while I found the first hour fairly interesting, it suddenly grew completely stale, didn't introduce anything new for the remaining time, and just copied & pasted the same sorting puzzles over and over again. It would probably be more entertaining if the devs removed half of the content. You also can't unlock Gold right away in some puzzles because you need to unlock hardware upgrades for your computer first. Ugh.

This game is barely, tenuously about machine learning. The first few puzzles are solved with if/else or if/elseif statements that this game grandly calls Expert Systems and Decision Trees. The puzzles at least early on are mostly figuring out how to route red/green/blue circle/square/triangle to the correct output, on a time limit for a gold/silver/bronze medal.

Unfortunately even in some early puzzles the gold medal result can only be achieved by RNG. The items to be routed appear to come out in a fixed, mixed order. When routing an item that matches neither case of an if/elseif block it will 'randomly' go to either of the outputs of the block, except that the randomness is decided when you edit your layout, and rerunning the scenario won't change the result.

I could not get a gold time on one puzzle with the obvious solution after several reruns and realised that the items were being routed exactly the same every run despite the tutorial claiming randomness. Messed around with different layouts still not getting the gold time, then reverted back to the original solution and suddenly met the time just because the random routing resolved differently after a rebuild.

This is just poor design that is only exacerbated when the game introduces 'perceptrons' that have a fail-rate for their routing. You can, at best, implement the solution 'most likely' to succeed and close, reopen, and rerun the puzzle to force the randomness to be decided again until you meet the requirements.

Having to coincidentally succeed at a puzzle game is just bad.

Sadly not really that enjoyable. It's not a coding game. The optimization part often seems rather random to fit the mechanics instead of making sense. The special levels (e.g. the genetic algorithm levels) simply don't work for me, as no matter what I do, the AI doesn't change it's output.