AI for human performance analytics

Our Clinical AI research focuses on developing explainable, video-based AI systems for clinical procedural skill assessment, workforce development, and lifelong improvement. The central goal is to move beyond traditional satisfactory/unsatisfactory checkoffs and better understand the “grey area” of human performance—whether a procedure was completed correctly, how effectively it was performed, and what contextual factors influenced the result. To support this work, we are developing the CLARA Dataset, collected from authentic clinical skill-validation activities at Purdue. It currently includes 174 student validation videos across nine procedures, involving approximately 200 students and 18 instructors, together with instructional videos, instructor review videos, and expert-curated error knowledge.

Our framework combines temporal video analysis and vision-language models to evaluate procedural performance across correctness, performance quality, and contextual allowance or difficulty, while also capturing safety-critical errors, expert disagreement, and uncertainty.