Yotam Mann, Google Creative Lab

A user plays a few keys on a piano and an RNN model generates more according to your input