LLM-as-Judge in Education: A Curriculum-Grounded Marking Pipeline

The paper introduces a curriculum-grounded LLM-as-Judge pipeline designed for automated question-level marking in educational settings, particularly for university admissions. This system uses a staged workflow to generate question-specific rubrics and evaluate student responses against established curriculum standards, enhancing consistency and transparency in marking. Preliminary evaluations indicate that the pipeline's marking outcomes are comparable to those of human tutors, with justifications that are more traceable to official educational artefacts, making it a significant advancement for practitioners in educational technology and assessment.