The application of Generative Artificial Intelligence (GenAI) to education is set to transform personalized learning by offering adaptive, real-time tutoring according to the needs of learners. In this work, we outline an architecture for a GenAI virtual teaching assistant (VTA) based on large language models (LLMs), retrieval-augmented generation, knowledge tracing, and multi-modal content generation. The proposed GenAI-VTA architecture combines three main components: a knowledge tracing module based on deep knowledge tracing (DKT), an answer generation module powered by an LLM along with knowledge retrieval from a knowledge base aligned to the curriculum, and finally a learning analytics dashboard for teachers. Performance analysis based on a controlled experiment involving 150 undergraduate students reveals that the use of the GenAI-VTA increases learning performance by 28.7%, lowers average response time below 1.5 seconds, and provides satisfaction levels up to 86%.
Pavithra et al. (Thu,) studied this question.