01 /
Project Description
The Virtual Teaching Assistant (VTA) project develops and evaluates a multimodal AI learning system that provides students with timely, trustworthy, and course-grounded support while helping instructors better understand student learning. Many existing AI assistants rely primarily on text and may struggle to interpret the diverse resources students encounter in STEM courses, including lecture slides, diagrams, figures, assignments, audio, and video.
The VTA addresses this challenge through a Mixture-of-Experts (MoE) architecture that integrates specialized text, vision–language, and audio capabilities. Student questions are dynamically routed to appropriate AI experts, while responses are grounded in course materials and linked back to their sources. At the same time, the system provides instructors with learning analytics that surface common misconceptions, engagement patterns, and areas where students may need additional support. The project will pilot the VTA in an undergraduate classroom to examine both its technical performance and its potential to support student learning and instructor decision-making.
WSU Insider: NVIDIA grant will support AI for teaching and learning
02 /
Project Vision
AI learning assistants should do more than answer students’ questions—they should understand the multimodal contexts in which learning occurs, provide responses grounded in trustworthy course evidence, and generate insights that teachers can use to improve instruction.
The VTA project is designed to connect multimodal AI and learning analytics within a unified human-centered system: providing students with just-in-time learning support while helping instructors recognize patterns in student understanding and respond more effectively to emerging learning needs.
03 /
Project Goals
Develop a trustworthy multimodal Virtual Teaching Assistant
Develop a source-grounded AI system that integrates text, visual, and audio information to provide accurate and contextually relevant support for student learning.
Transform multimodal learning data into actionable instructional insights
Develop instructor-facing learning analytics that identify common misconceptions, engagement patterns, and emerging learning needs to support instructional decision-making.
Evaluate the VTA in authentic learning environments
Pilot the system in an undergraduate course to examine technical performance, student engagement and learning, instructor workload, usability, and opportunities for future scaling.
04 /
Team & Collaborators
Project Team
- Parteek Kumar
- Peng He
Collaborators
- Washington State University School of Electrical Engineering and Computer Science
- College of Education, Sport, and Human Sciences
- NVIDIA
Partners
- Washington State University undergraduate courses and instructional programs