Research

My research and my work is creating effective and engaging products using artificial intelligence. I research best practices, lead projects, design innovative interfaces, model the user, develop prototypes, write code, conduct user testing, data mine for continuous data-driven improvement, and write papers describing advances in the state of the art.

I am currently working with Carolyn Rosé at the Teledia Lab in Carnegie Mellon University's Language Technologies Institute. Our primary goal is facilitating collaboration between people and between people and computers. To that end, we employ natural language processing, machine learning, and principles of human-computer interaction. One of our exciting new projects is a virtual tutor that will take as input students' spoken language, facial expressions, physical location, and body position, and respond with spoken dialog along with appropriate facial expressions, gaze direction (to look at a particular student based upon relative physical location), and body gestures. Another exciting project is facilitating academically productive interaction among students engaged in "Mob Programming" in a cloud computing course.

Previously, I worked as a Cognitive Scientist for Carnegie Learning, designing and developing a large proportion of the company's Cognitive Tutors, mainly for math and also for English and financial literacy. Cognitive Tutors are intelligent tutoring systems (software) that are widely recognized as best in class for delivering proven, effective instruction. They have been used by millions of students in thousands of schools.

I earned my PhD in Intelligent Systems at the University of Pittsburgh. My advisor was Kurt VanLehn. I worked at the Learning Research and Development Center, a multidisciplinary research center for understanding and improving learning.

As part of my PhD work, I developed DT Tutor, a decision-theoretic software engine for selecting the actions that an intelligent tutoring system will take. This approach, embodied in a dynamic decision network, ensures that DT Tutor's actions are optimal given its beliefs and objectives. DT Tutor considers a rich set of tutorial state attributes in a quest to approach some of the sensitivity and subtlety exhibited by human tutors. These attributes include the discourse state, progress on the tutorial task (e.g., solving problems), and the student's knowledge, focus of attention, and affective state.