The written examination remains a widely used tool for thoroughly assessing all student performance in education. It gives teachers a typical way to guarantee things are the same and to stop prejudice in grading. A teacher's experience, comprehension of real ideas, grading policies and due dates affect all common ways of measuring students. These processes consume a great deal of everyone's time, too. Because typical evaluation methods are expensive, lengthy and resource- intensive, the process of getting and verifying outcomes often slows down. The AI-Based Paper Evaluation System helps make grading exam answers automatic, improving its efficiency, accuracy, as well as the amount that can be graded. The system systematically sorts answers from scanned PDF answer sheets by sections (A, B and C) and specific questions (such as Q1a and Q1b) for full assessment. It leverages the Google Vision API to extract text from handwritten student papers, ensuring accurate data capture. The Cosine similarity, Jaccard similarity and BERT-based NLP model assigns similarity ratings, each of which represents total alignment with the predicted result; this model assesses how similar each of the student responses, in addition to model answers, are. Each section undergoes an evaluation. The evaluation is independent. The similarity scores are added to produce the section grades and overall grades.