you did exactly what the first robot's model actually looks for! The inference is firstly done with a regular "swipe to type" model, and the second one adds another inference layer on it, trained on how the cat actually walks the keyboard :)
