Abstract:

Within the framework of continuous assessment, frequent evaluation is essential to track student progress; however, the current higher education landscape is increasingly characterized by overcrowding and a significant rise in faculty workload. While detailed, personalized feedback -moving beyond a mere numerical grade- is a powerful pedagogical tool, providing it at scale has become practically unfeasible due to limited institutional resources. This creates a "feedback gap" where students receive a grade but lack the qualitative rationale needed to identify their strengths and weaknesses.

To address this challenge, this study evaluates the efficacy of Large Language Models (LLMs) in automating the generation of high-quality, personalized feedback reports. Unlike traditional automated grading systems, LLMs offer the ability to interpret complex human language and follow nuanced instructions. This research presents the development of an AI-driven assistant specifically designed for the academic environment of the Universitat Politècnica de València. The system is capable of processing digitized assessment samples, including both digital academic assignments and scanned handwritten exams.

The methodology integrates a structured rubric that contains the correct solutions and a set of pedagogical instructions. By mapping student responses against these criteria, the assistant automatically generates a comprehensive report for each student. This report justifies the assigned grade, provides a pedagogical explanation of errors, and suggests specific actionable measures to improve subject mastery.

The implementation results demonstrate that the AI-generated responses meet the rigorous quality and technical standards required for university-level engineering education. Furthermore, the tool significantly enhances student support by providing immediate, individualized mentorship. Ultimately, this work proposes a scalable solution that alleviates the administrative burden on teaching staff while restoring the formative value of assessment in massified educational settings.

Keywords:

Academic work, large language models, automatic report generation.