Programming exercises are a central instrument in computer science education, but their quality depends on consistent alignment between the problem statement, template, solution, and tests. Artemis already uses large language models (LLMs) to detect inconsistencies, yet the results appear as raw JSON that is hard to interpret and act on, which makes review slow and error-prone for instructors.
This thesis extends Artemis with AI-assisted review tools that make consistency checks accessible and actionable. It introduces a persistent review comment system, and consistency issues are represented as review comments within that system. The exercise editor visualizes these issue comments inline with severity, affected locations, and suggested fixes, and provides an overview for fast navigation. Instructors can preview and apply proposed corrections directly in the editor while keeping human control over final changes. These contributions improve the reliability of programming exercises and reduce manual overhead.
Methodologically, the work follows a design-oriented software engineering approach. It derives requirements from instructor workflows, maps them into a system design with persistent issue storage and editor integration, and implements the system within Artemis.