Stroke is a chronic, lifelong illness, and full recovery requires continuous physical and cognitive rehabilitation. Such long-term rehabilitation is cost-prohibitive; however an approach to providing long-term therapy that has recently gained traction is the use of socially assistive agent (SAA) systems. These systems make use of non-contact communication devices and can be used to guide people through a variety of rehabilitative tasks. They have the potential supplement current rehabilitation practices by providing motivation during intense exercises, and can extend the reach of the therapist into remote and home settings. Though SAA systems have been used in a variety of rehabilitative and assistive contexts, there remain questions regarding the best design for such systems. Currently there is a lack of detail on what type of feedback optimizes user performance, and the role that the delivery medium (e.g., a human coach, a tablet, or a robot) plays in user performance. The purpose of this thesis is the design of a system to investigate the interaction between feedback and medium type when implemented for a novel motor task. The selected task is modeled on the shuffleboard game, with the delivery medium including a human coach and tablet and two types of augmented feedback. The designed system incorporates various hardware and software components. A vision system communicates with a laptop to record and analyze motor task data, with a program that also interfaces with a control circuit. The control circuit may transmit data through Bluetooth to a custom-built app on the tablet, which then provides augmented feedback with audio dialogue. Otherwise, the human coach is provided designed feedback from the laptop. An initial system evaluation was performed with this constructed system using pilot participants to validate the design. The initial system evaluation demonstrated the ability to improve participant performance; however, it also demonstrated a high level of task difficulty. Several changes may need to be incorporated to the system to ensure better learning for participants. This includes changes to the physical setup, as well as changes to the frequency of the augmented feedback. This thesis may be used as the foundation for future experimentation with different delivery media or types of augmented feedback to discover how to best optimize user performance for a novel motor task.