Wow, really cool! Could you apply this to the behavior dataset that we have at Amplitude? I'd love to talk more.