In this demo, we will showcase “Millie” our live digital fitness trainer. She is an on-screen avatar that interacts in real-time to observe and grade the user’s movements for speed, accuracy and form. Millie is powered by an audio-visual neural network that processes the video and voice input using 3D convolutions to power the avatar’s responses. The neural network is trained on several million sample videos showing crowd-workers performing actions that they recorded using consumer grade RGB cameras and microphones. This allows the digital fitness trainer to run on a wide range of hardware devices, such as iPhone, iPad and in the cloud.