Hybrid Human-AI Framework for Personalized Feedback and Student Learning Enhancement via Exercise Assessment in Programming Education

Effective learning requires timely identification of individual student strengths and weaknesses to support targeted formative feedback. However, human only assessment is often limited by scalability and consistency, while fully automated systems may lack pedagogical transparency and contextual understanding. This paper presents a hybrid human AI framework for assessing student exercises that integrate machine learning based analysis with structured human oversight to enhance personalized feedback and learning outcomes. The proposed framework analyzes student generated data, including programming code submissions and short textual explanations, to construct individualized assessment profiles that guide learning progression. Automated techniques based on natural language processing and code analysis provide scalable evaluation, while human evaluators validate results and address cases requiring nuanced pedagogical judgment. The framework is evaluated using a dataset of 1,200 undergraduate programming exercise submissions collected from a Java programming course. A comparative analysis of human only, AI only, and hybrid assessment approaches is conducted to examine differences in consistency, fairness, and feedback quality. The results indicate that hybrid assessment outperforms single modality evaluation by improving the reliability and actionable value of feedback provided to students. The findings suggest that structured human AI collaboration represents an effective approach for personalized assessment in programming education, supporting scalable evaluation while preserving pedagogical integrity.