Automated ENEM essay scoring and feedbacks: a prompt-driven LLM approach

Abstract

This paper presents a novel LLM-based solution for automated evaluation of essays from the National High School Exam (ENEM), the largest educational test in Brazil, addressing the significant gap in Portuguese-language automated essay scoring (AES). Leveraging a chain of meticulously crafted prompts and advanced LLM architectures, our system achieves 100% adjacent agreement with human raters across all five ENEM competencies and the final score, significantly outperforming existing methods. Beyond accurate scoring, the system generates detailed, competency-specific feedback, transforming it into a valuable learning tool for students. Our approach, based on prompt engineering and adhering strictly to official ENEM scoring rubrics, offers a robust and scalable solution for large-scale educational assessment in Portuguese, enhancing both efficiency and educational value.