<![CDATA[Over the years, balancing personalized support with learner motivation has proven difficult in programming education. Traditional instructional approaches seldom adapt to individual needs, often resulting in uneven engagement and cognitive overload. Gamified settings can enhance motivation, but many lack adaptive mechanisms informed by motivational frameworks such as Self-Determination Theory (SDT). As a result, few systems combine reinforcement learning (RL) with probabilistic learner modelling to deliver scalable, personalized learning pathways. This paper proposes DevQuizzer, an adaptive game-based learning (AGBL) system designed to support personalized programming instruction. The system combines RL and Bayesian Networks (BNs) models to dynamically adjust task difficulty, feedback, and learning pathways according to learner performance and engagement. A dual-mode instructional design: code-first and theory-first, accommodates various learning preferences, while gamified elements sustain motivation. We evaluated AGBL with students from five universities using a mixed-methods approach that combined quantitative measures of usability, engagement, and adaptivity with qualitative insights into learner experiences. Results confirmed the system’s reliability and construct validity, with learners reporting high satisfaction, autonomy, and perceived learning gains. The findings show the potential of hybrid RL-BN models to provide transparent, interpretable, and scalable adaptive learning environments for programming education.]]>
