In the FLINK project (“Feedback-Based Learning Support in Computer Science Studies”), the Department of Computer Science at RPTU Kaiserslautern is developing innovative approaches to provide individualized support to students in programming during the early stages of their studies. The starting point is the challenge that students often begin their studies with highly varied levels of prior programming knowledge. This leads to a performance gap that persists even after introductory programming courses (Bui et al., 2023). The project therefore aims to bridge these knowledge gaps through tailored, learning-promoting feedback. The project is based on two central approaches:
- AI-powered feedback: Programming submissions are automatically analyzed on the established learning platform ExClaim. Based on extensive exercise data, the system generates personalized feedback that goes beyond traditional error messages and provides concrete insights into the causes of errors as well as next learning steps. Feedback can be generated even before the submission deadline, thereby supporting students’ learning process while they are still working on the assignment.
- Eye tracking: We are also investigating cognitive barriers encountered while working on programming tasks using physiological markers such as eye tracking. The goal is to determine the ideal time to provide assistance and thus resolve learning obstacles early on.
Unlike conventional chatbots, our system does not provide ready-made model solutions but, based on the principle of cognitive apprenticeship, encourages independent reflection and problem-solving. FLINK thus not only promotes subject-specific competence but also specifically strengthens students’ metacognitive strategic knowledge. The project thus serves as a model for data-driven, personalized, and scalable learning support in modern higher education.
The ExClaim platform is being expanded so that the system can be easily adapted to other programming languages and other text-based assignments. The possibility of adapting it to non-text-based assignments, such as diagrams or images, is also being explored.
The entire pipeline is operated using a server infrastructure that complies with data protection regulations. In addition, we use only so-called "open weight " language models. This means that all data remains in a secure environment. This allows us to ensure that feedback generation is conducted in a manner that is both secure and compliant with data protection regulations.
Evaluation
The project is accompanied by a formative evaluation involving students. The findings are directly incorporated into the optimization of the assignments and feedback. Initial insights were gained during a pilot study in the summer semester of 2026 as part of the “Programming and Modeling” course. The resulting optimizations will be evaluated during the second evaluation phase in the winter semester of 2026/27, in conjunction with the “Programming in C” course. During this phase, details of the pipeline will be adjusted, and other language models will also be tested.