International Journal of Science, Engineering and TechnologyAuthors: H V Poojitha, Harshitha C, Madhulekha M, Likhitha B
Abstract: Evaluating academic assignments manually requires considerable time and effort, particularly when an instructor must assess many student submissions against common questions and evaluation criteria. Differences in interpretation may also lead to variation in the feedback provided to students. This paper presents an AI-assisted academic assignment evaluation system that supports teachers in reviewing written student responses. The system accepts questions, reference answers, and evaluation rubrics from a teacher, extracts text from submitted assignment files and compares student responses with the corresponding reference material. A pretrained Sentence-BERT model is used to generate semantic sentence representations, and cosine similarity is applied to estimate the semantic closeness between the student's response and the reference answer. The resulting similarity information is considered within a rubric-oriented scoring workflow, while rule-based checks support the evaluation process. The application is implemented using a React/Vite frontend, a Python FastAPI backend, and an SQLite database. Sentence-BERT is used in it pretrained form without task-specific fine-tuning. The completed testing reported for this project focuses on functional verification of the application workflow; a benchmark against human-assigned marks and quantitative grading-accuracy results have not been established. The system is therefore intended as a decision-support tool that can reduce repetitive evaluation effort and provide structured feedback, with the teacher retaining responsibility for final academic judgement.
DOI:
