Enhancing University Students’ Activities with Interactive Immediate Feedback Using Customized LLMs

One of the significant issues for the students is getting timely feedback on their work so that they can understand their mistakes and improve their knowledge iteratively. There have been significant results in this area through automated evaluation, but they require a strongly defined structure of the expected results, which has its limitations. We propose using customized large language models as the core of an assignment evaluation software architecture that is used for automated evaluation and feedback generation on a variety of tasks in an introductory Java programming course. The model is implemented and tested, achieving significantly better results than using general-purpose models.