This video gives an overview of EMPI, the embodied musical predictive interface. This interface allows constrained call-and-response interaction with a musical machine-learning model. The performer can control a basic synth sound with one lever. The machine learning model responds with the same sound and another lever controlled by a servo. We tested the EMPI with three ML models: one based on human-sourced data, as well as a synthetic dataset, and a noise dataset as a control. We also tested the EMPI with the motorized lever enabled and disabled. This video shows example interactions with each of the six conditions.