Investigates how real-time multimodal context improves LLM-based AR task guidance assistants. Develops a context injection framework that progressively incorporates task descriptions, procedural steps, real-time task progress, and hand action data into an LLM's prompt. Validated on the HoloAssist dataset; real-time hand action data provided the largest performance gains in accuracy, relevance, and consistency.

A gamified VR training system for automotive assembly workers combining high-fidelity simulation with interactive game mechanics. Features synchronized multimodal feedback (visual, auditory, haptic) via headset and controllers to study each sensory channel's role in skill acquisition and transfer. Optimized for wireless headsets (Meta Quest 3) with cloud-based analytics dashboards for progress tracking and longitudinal skill assessment. Demonstrated in the context of Kia's Door Module Assembly using a "Ghost Master" scaffold grounded in Cognitive Apprenticeship theory.

AR-guided psychomotor tasks using HoloLens 2 for precision inspection with 3D-printed parts and industrial gauges. Two-session study covering gauge selection, calibration, measurement of part features, and accept/reject decisions based on tolerances.