AI-assisted assessment of software engineering coursework: Tool-support and an empirical study

Context Grading coursework at scale is a persistent challenge in software engineering and computer science education, with issues of time constraints, grading consistency, and the need for timely, high-quality feedback. Objectives Using a BSc final-year software testing course as context, this study investigates ChatGPT's effectiveness in assisting the grading of practical lab reports and Java test-automation code, assessing grade alignment with human evaluations, feedback quality, and students' perceptions. Methods We developed AIAssessor, a research instrument for studying AI-assisted grading workflows, to batch-process student submissions via prompt-engineered ChatGPT API requests. Based on the tool and the approach, the study also presents a reusable framework named Instructor-Led and Instructor-Calibrated AI-assisted Assessment (ILCAA) , synthesizing four transferable principles: rubric anchoring, iterative prompt calibration, structured feedback generation, and instructor oversight. Results In this preliminary exploration, prompt refinement led to improved grade alignment in most labs, though findings are illustrative given small sample sizes. AI-generated feedback was generally clear and actionable, though occasional hallucinations reinforced the need for instructor review. Surveyed students valued AI feedback but emphasized that final grading responsibility must remain with instructors. Conclusion ChatGPT can serve as a practical grading assistant under a human-in-the-loop model. Responsible adoption requires careful prompt design, iterative calibration, and sustained instructor oversight to ensure pedagogical integrity and student trust.