AI-Assisted Continuous Assessment in Programming Education: Design and Evaluation

Automated assessment of programming tasks remains a persistent challenge in computer science education, especially when scalability, consistency, and formative feedback are required. This study presents the design and evaluation of an AI-assisted continuous assessment system that leverages Large Language Models to evaluate programming assignments, generate feedback, and support learning analytics. The system integrates rubric-based evaluation, prompt engineering, and human-in-the-loop validation to ensure both reliability and transparency. A mixed-method experiment was conducted in an undergraduate programming course involving over 1100 student submissions across multiple assignments and subjects. Results show a strong correlation (R2 = 0.87) between LLM-based and instructor traditional grading, with substantial reductions in assessment time (over 75%) and increased consistency. Qualitative feedback from instructors and students highlighted the ability of the tool to provide immediate, explainable, and unbiased feedback, improving engagement and learning outcomes. Despite these advantages, limitations include occasional misclassifications and the need for pedagogical oversight. The findings indicate that LLM-assisted systems can significantly enhance scalable and equitable continuous assessment in programming education, provided that ethical use and human validation remain integral components.