Abstract

This paper proposes an AI-based programming tutoring system that assists learners in detecting programming errors, understanding their causes, and iteratively improving their code. Instead of simply providing corrected code, the system employs a learning cycle comprising task definition, code and test-case submission, error analysis, evidence-based explanations, targeted hints, and learner verification. Its architecture integrates code-structure analysis, test execution, and a language model for feedback generation, together with a control layer designed to minimize inappropriate or unsupported AI suggestions. The paper presents the theoretical foundations, system architecture, operational workflow, feedback model, interface prototype, an illustrative Python programming scenario, and a proposed experimental design. As no classroom experiment has yet been conducted, the findings are limited to system design and illustration rather than quantitative evidence of learning effectiveness. The proposed architecture provides a foundation for future research on programming assessment, personalized feedback, and the responsible use of AI in information technology education.

Keywords: AI tutoring, automated feedback, code analysis, error detection, problem-solving competence, programming education.