Development of an AI-assisted Coding Tutor for Secondary Education

Thesis Abstract

The rapid integration of coding into secondary education has amplified the demand for scalable, personalized learning tools that can accommodate diverse learner profiles and pacing. Despite advances in educational technology, many classrooms lack adaptive scaffolding for novices, leading to persistent gaps in computational thinking, program comprehension, and problem-solving fluency. This study develops and evaluates an AI-assisted coding tutor designed to operate within standard secondary school curricula, offering real-time feedback, adaptive hinting, and formative assessment aligned with national standards in Computer Science Education. The aims are to (1) design an AI tutor that individualizes instruction through dynamic learner modeling, (2) implement pedagogically grounded coding activities spanning block-based and text-based paradigms, and (3) evaluate its impact on student achievement, engagement, and self-regulated learning strategies. A mixed-methods design was employed in a multi-site quasi-experimental study across three public secondary schools over a 12-week intervention. The population comprised 1,200 students aged 12–16 enrolled in introductory coding courses. A stratified random sample of 600 students was selected, with 300 assigned to the AI-assisted tutor condition and 300 to a traditional instruction control group. Data collection instruments included standardized pre- and post-tests measuring computational thinking and programming proficiency, the System Usability Scale (SUS) for perceived tool usability, the Motivated Strategies for Learning Questionnaire (MSLQ) for self-regulated learning constructs, and classroom observation rubrics to capture engagement. Instrument validity and reliability were established through content validation by a panel of five CS education experts (CVR > 0.78) and a pilot study (Cronbach’s alpha for major scales 0.84–0.92). Data were analyzed using a combination of inferential statistics and qualitative synthesis multiple regression to assess predictors of achievement, ANCOVA to compare post-test outcomes controlling for baseline scores, and MANCOVA for cross-domain effects (knowledge, debugging skills, and transferability). For qualitative data, thematic analysis was conducted on 36 teacher interviews and 180 student reflection entries to triangulate quantitative findings. Key expected findings include (1) significant gains in computational thinking scores and programming proficiency for students in the AI-assisted group compared with controls (p < .05), with effect sizes in the small-to-moderate range (Cohen’s d ? 0.35–0.50); (2) higher engagement indicators and greater use of metacognitive strategies among AI-tutor users, as evidenced by MSLQ subscale scores and in-situ observation data; (3) positive perceptions of usability and perceived usefulness of the tutor, with SUS scores averaging above 72, indicating acceptable to high usability; (4) differential benefits for students with lower prior achievement and for those requiring more structured scaffolding, suggesting the tutor’s adaptive features effectively close attainment gaps. Theoretical interpretation will be anchored in constructivist– sociocultural perspectives and the cognitive apprenticeship model, with the AI tutor operationalizing scaffolds such as fading, legitimate peripheral participation, and reflective practice. The study will also examine the role of the learner model in driving personalized feedback loops and hint strategies. The study contributes to knowledge by empirically validating an AI-driven, scalable instructional intervention that integrates adaptive feedback with authentic coding tasks within mainstream curricula, thereby addressing equity and scalability in CS education. It offers a validated methodological framework for evaluating AI-assisted tutors in secondary settings, identifies design features that maximize learning gains, and delineates conditions under which AI tutoring yields the greatest benefits. Practical implications include guidance for curriculum designers, teachers, and policymakers on implementing AI-supported coding instruction, considerations for data privacy and classroom integration, and recommendations for professional development focused on leveraging learner analytics. The main conclusion anticipated is that an AI-assisted coding tutor improves learning outcomes and engagement when embedded in a coherent pedagogical model with transparent feedback and alignment to classroom assessment practices; recommendations include iterative refinement of the learner model, expansion to project-based coding activities, and longitudinal studies to assess transfer to higher-level programming concepts.