Assessment is fundamental to programming education; however, it is a labour-intensive and complicated process, especially in extensive learning contexts where it relies significantly on human teachers. This paper presents an automated grading methodology designed to assess Python programming exercises, producing both continuous and discrete grades. The methodology incorporates GPT-4-Turbo, a robust large language model, and machine learning models selected by PyCaret’s automated process. The Extra Trees Regressor demonstrated superior performance in continuous grade prediction, with a Mean Absolute Error (MAE) of 4.43 out of 100 and an R² score of 0.83. The Random Forest Classifier attained the highest scores for discrete grade classification, achieving an accuracy of 91% and a Quadratic Weighted Kappa of 0.84, indicating substantial concordance with human-assigned categories. These findings underscore the promise of integrating LLMs and automated model selection to facilitate scalable, consistent, and equitable assessment in programming education, while substantially alleviating the workload on human evaluators.

Accurate estimation of learners’ skill mastery is essential for adaptive learning systems, particularly in programming education, where evaluating learner performance remains challenging due to the complexity of code assessment and the need to capture partial correctness and diverse learning behaviors. This study aims to improve mastery estimation and adaptive exercise recommendations through a continuous knowledge-tracing approach based on an extension of Bayesian Knowledge Tracing. The proposed approach integrates continuous performance scores from an automated evaluation process that combines structural code analysis, dynamic testing, and large language model–based analysis, while accounting for task difficulty and guidance format. The estimated mastery is then used to recommend programming exercises adapted to the learner’s knowledge state. To evaluate the approach, controlled simulations involving diverse learner profiles and real interaction data were conducted. Results show that the proposed model variants reduce mastery estimation error by up to 70% compared with the baseline model, particularly for learners with low and average proficiency levels. Furthermore, the approach improves estimation stability and generates recommendations consistent with learner progression. The main contribution of this work is a continuous extension of Bayesian Knowledge Tracing that leverages fine-grained assessment data to improve learner modeling and adaptive exercise recommendation in programming education.