Manual evaluation of student assignments remains a significant bottleneck in large engineering courses, where instructors often grade hundreds of submissions under tight deadlines
Abstract & Keywords
Manual evaluation of student assignments remains a significant bottleneck in large engineering courses, where instructors often grade hundreds of submissions under tight deadlines. This paper presents an automated assignment evaluation framework that combines large language model (LLM) based rubric scoring with embedding-based plagiarism detection to assist instructors rather than replace them. The system ingests student submissions in PDF and DOCX formats, extracts structured content, and scores each submission against a instructor-defined rubric using a retrieval-augmented prompting strategy. A parallel plagiarism module compares submissions pairwise using sentence-transformer embeddings and flags suspicious overlaps for human review. We evaluate the system on a corpus of 240 undergraduate assignments across two courses and compare automated scores against grades assigned independently by two teaching assistants. Results show a strong correlation between automated and human scores (Pearson r = 0.89) while reducing average grading time per submission by more than 75%. The plagiarism module achieves a false-negative rate of 4.7%, a substantial improvement over keyword-based baseline tools. We discuss deployment considerations, instructor override mechanisms, and limitations related to open-ended and creative assignments.
Author Affiliations
References Listing (3)
- [1]K. R. Rao and J. Doe, "High-resolution convolutional modeling in thermal imaging systems," IEEE Trans. Image Process., vol. 31, pp. 210–222, Jan. 2024.
- [2]W. Liu, R. Chen, and M. Rossi, "Deep residual neural networks for noise-reduction and super-resolution edge rebuilding," Pattern Recognition, vol. 142, pp. 110-125, Oct. 2025.
- [3]S. Jenkins, "Cooperative token-heuristics inside warehousing routing grids," Int. J. Rob. Res., vol. 18, no. 4, pp. 450–467, May 2025.