"AutoGrade": an AI-Based Assessment Tool for Computer Science 1

Assessment is a crucial element of the educational process. This study explores the use of AI to assist in grading midterms for the Computer Science I course. Currently, midterms are conducted in person, on paper, closed book, and are graded using detailed rubrics. With an increasing number of students and teachers, an AI tool for an initial grading could enhance speed and accuracy for all teachers. The goal is to achieve preliminary grading according to a rubric, which a teacher then validates. This paper details the course and the process, including an application that facilitates uploading photos of students’ tests, transcribing images, allows uploading a rubric, and applying it. In the experiment conducted, 24 student exercises were considered, using a rubric with 7 criteria. Results show an 85% agreement between the application of the rubric by AI and the teacher (143 out of 168 criterion). Recommendations and conclusions are provided.