Evaluating handwritten answer scripts by hand has long been a resource-intensive process, prone to inconsistency and examiner fatigue. As student enrolments continue to rise across educational institutions, the growing volume of answer sheets presents a serious logistical challenge that traditional grading methods are not equipped to handle efficiently. Recent developments in Artificial Intelligence, Machine Learning (ML), Optical Character Recognition (OCR), and Natural Language Processing (NLP) offer a technically credible path toward automating this workflow.

This paper presents an ML-driven system designed to evaluate handwritten answer sheets without manual intervention. The pipeline begins with OCR-based conversion of scanned scripts into machine-readable text, followed by NLP preprocessing and multi-strategy scoring. Evaluation draws on keyword matching, semantic similarity, and syntactic analysis, with transformer-based models—BERT and Sentence-BERT—paired with cosine similarity and Jaccard index metrics to achieve greater precision.

The resulting system produces consistent, bias-free scores, generates detailed performance reports for individual students, and can operate at scale across diverse exam formats, file types, and languages. By uniting document recognition with semantic understanding, the proposed solution addresses key shortcomings in fairness, turnaround time, and transparency that persist in conventional grading practice.

Keywords— Automated Grading, Optical Character Recognition, Natural Language Processing, Semantic Similarity, Machine Learning, Transformer Models