A LARGE LANGUAGE MODEL-BASED FRAMEWORK FOR ENHANCING PERSONALIZED LEARNING OF C PROGRAMMING

Abstract

The increasing demand for competent programmers has amplified the need for instructional approaches that are both effective and personalized. Traditional programming education methods often fall short in offering real-time feedback, adaptive task difficulty, and contextual support, challenges that limit learner engagement and hinder skill acquisition. While recent advances in Generative Artificial Intelligence (GenAI), particularly through Large Language Models (LLMs), offer potential to address these issues, existing educational frameworks rarely integrate such tools in a way that ensures accuracy, coherence, adaptability, and personalized feedback. This study addresses this gap by developing and evaluating a framework that operationalizes GenAI for personalized C programming instruction among undergraduate students. A systematic literature review was conducted to identify key features essential for effective AI-enhanced learning, including adaptability, contextual relevance, coherence, accuracy, prompt engineering, personalized feedback, creativity, and usability. These features were validated through expert opinion and used to guide the development of a web-based learning framework. The framework integrates four key components: an LLM interaction module (Node.js/Express.js), an adaptive learning engine, a collaborative coding platform, and a React.js interface to deliver real-time, learner-specific support in C programming. Architecture design followed agile prototyping cycles with iterative expert feedback. To evaluate effectiveness, a mixed-method study was conducted with 139 undergraduate programming students at Murang’a University of Technology. A purposive sample of 75 formed the experimental group and engaged with the GenAI powered system over two weeks, while 64 students in the control group received conventional instruction. Data collection involved structured questionnaires, open-ended surveys, and a pretest-posttest design aligned to core programming concepts. Performance was assessed through gain scores in pretest-posttest evaluations, mean ratings on personalization features, and learner feedback on usability and engagement. The findings showed high student satisfaction, with a mean of 4.13 for usability and 4.14 for perceived learning enhancement. A pretest-posttest comparison revealed a mean gain of 9.08 points in the experimental group, versus 2.14 in the control group. A Mann-Whitney U test confirmed this difference was statistically significant (p = 0.008). This had a real impact, as more than half of the students (55%) said the framework helped them become better at solving programming problems. These results demonstrate that LLM-based tools can enhance personalized programming education and have strong potential for industry use in developer training and reskilling. The study recommends further refinement and ethical integration in academic settings. Cross-sector collaboration is also encouraged to support responsible, scalable adoption.