Automated evaluation of handwritten academic scripts remains a challenging problem due to variability in handwriting styles and the complexity of subjective grading. This paper presents ASEAS (Automated Script Evaluation and Analysis System), a multi-stage AI-based framework that integrates Transformer-based Optical Character Recognition (OCR), semantic embedding models, Retrieval-Augmented Generation (RAG), and rubric-constrained Large Language Models for automated academic assessment. The proposed system first performs high-accuracy handwritten text recognition using a fine-tuned Transformer OCR module. The extracted text is then processed through a retrieval-enhanced semantic evaluation pipeline that aligns responses with rubric expectations and reference content. Experimental analysis on pilot-scale assignment datasets demonstrates transcription accuracy exceeding 94% and strong agreement with human evaluators (overall Pearson correlation ≈ 0.87). Ablation studies further confirm the importance of contextual retrieval and rubric grounding in improving grading consistency. The results indicate that ASEAS provides a scalable, consistent, and time-efficient alternative to manual grading, offering approximately threefold reduction in evaluation time while maintaining strong semantic alignment with human assessment standards.
M et al. (Thu,) studied this question.