Gerridae is a sound toy heavily inspired by Electroplankton.
Due to loose training, the ML agents don't behave exactly in the way that I had originally planned. While I intended for them to follow their targets, the result I got is that they kind of do, but are mostly bad at it. For a sound toy, I think this result is actually more interesting than if they strictly followed their targets because it produces more variation without feeling random.
I'm honestly not particularly interested in ML and AI tools, but when I do use them, I think there's a lot of value in embracing their finickiness and imperfections.