Automating the pre-submission review of academic reports via the integration of formal rules and large language models

Abstract

The article examines the challenges of delivering high-quality feedback on structured academic reports in higher education. It considers the problems with manual pre-submission reviews and the limitations of existing automated writing evaluation tools in performing semantic analysis. During the research, a Human-Aligned Rubric and Review System (HARRS) with a hybrid architecture combining rule-based logic with constrained generative AI was developed. The impact of anchoring LLMs within human-defined rubrics to mitigate scoring bias and inconsistency is discussed. The practical aspects of implementing the HARRS pipeline and its performance relative to expert human judgment are described. The study shows that this hybrid approach maintains disciplinary standards while providing a reliable mechanism for enhancing student self-regulation.