vTA: How an Instructor Leverages Large Language Models for Superior Student Learning
Vivek K. Pallipuram, Vineeth Sai Varikuntla, Abdullah Tariq Choudhry · 2025
The increasing availability of large language models (LLMs) including ChatGPT-4o and Google’s Gemini have made valuable information accessible to students. While these generative artificial intelligence (gen-AI) tools can revolutionize the education landscape, they also bring threats and weaknesses. Their ability to effectively generate text outputs that approximate a human response raises ethical concerns regarding students’ work and learning. This issue, exacerbated by the loss of data privacy, fuels the skepticism surrounding the use of LLMs in education. This book chapter aims to alleviate those concerns and enable educators to embrace gen-AI for enhanced teaching and learning. We present a private, user-friendly software framework called virtual teaching assistant (vTA), which allows instructors to leverage their personalized LLM to aid student learning. The vTA’s core comprises a local LLM that undergoes three phases before facing students. The three phases include initial-tuning, domain-specific fine-tuning, and deployment. In the initial-tuning phase, the instructor uses prompt engineering techniques to generate initial specifications. This process tunes the LLM to meet the professor’s immediate expectations. In the fine-tuning phase, vTA and the instructor conduct interactive, active prompting to further refine vTA’s. In the deployment phase, the framework equips instructors with techniques to deploy their trained vTAs for teaching. We demonstrate vTA using the authors’ upper division/graduate engineering course, digital image processing, which includes advanced mathematical and programming concepts. The final goal of this chapter is to make LLMs accessible to instructors regardless of their technical expertise and facilitate effective and ethical learning.