The real-time, interactive feedback system we developed, Sensei: The T'ai Chi Teacher, is presented. This system provides a good platform on which to build a more sophisticated teaching, training, and feedback tool for gestures or action. In this document, we hope to substantiate that thesis by showing that Sensei does have the components necessary to be a good foundation. First of all, it is capable of performing real-time, multiple user, gesture recognition. Users of the system are free to practice T'ai Chi gestures, starting anywhere in the sequence of moves, and the system recognizes their actions. In addition, a complete teaching system must be able to give both positive and critical feedback to the user. This ability implies a knowledge of the instants in the user's performance of a gesture where the user was both least and most accurate in the movement. Both tasks are accomplished through the use of Hidden Markov Models. Experiments testing these abilities are presented. The work concludes with a discussion of future extensions.