Large Language Models for Personalized and Scalable Computer Education

Abstract

Large Language Models (LLMs) are increasingly used in computing education, offering new opportunities to support programming practice, generate explanations, and scaffold student learning. However, current LLMs are not pedagogically aligned and often fail to adapt to students’ prior knowledge, leading to explanations and feedback that are either too advanced or oversimplified. This research addresses the gap by developing human-centered LLMs tailored for programming education. Building on prior work in knowledge tracing and knowledge component discovery, we explore how student data (e.g., submissions, process data, misconceptions) can be leveraged to personalize LLM outputs. Our UKICER 2025 study demonstrated that automated knowledge component extraction with LLMs can approximate expert-annotated learning curves, highlighting the potential of scalable student modeling. The ongoing PhD project extends this by investigating prompt engineering, fine-tuning, and reinforcement learning with human or AI feedback (RLHF/RLAIF) to design pedagogically effective LLMs. The contributions include: (1) methods for tailoring LLMs to student profiles, (2) integration of automated KC pipelines into large-scale courses, and (3) evaluation of impacts on student learning, engagement, and experience. This work aims to advance scalable, personalized support in programming education through next-generation intelligent tutoring powered by LLMs.