Smart Paper Eval: An Intelligent AI-Based Automated Exam Paper Evaluation System

Abstract

The evaluation of handwritten answer sheets is a fundamental component of academic assessment, yet it remains a time-consuming, labor-intensive, and subjective process in traditional education systems. Manual grading is often prone to inconsistency, human bias, and limited scalability, particularly in large-scale examinations. This article presents Smart Paper Eval AI, an automated system utilizing Optical Character Recognition (OCR) and Natural Language Processing (NLP) for effective and impartial assessment of handwritten responses to address these challenges. Using a structured pipeline, the suggested system processes handwritten answer sheets in image or PDF format. To improve the input quality, first noise reduction and normalization are used as preprocessing techniques. To precisely retrieve textual material from handwritten inputs, a Transformer-based OCR model (TrOCR) is applied. Using Sentence-BERT, the retrieved material is then translated into semantic embeddings, hence allowing significant representation of the material. Semantic similarity analysis was then carried out by contrasting student responses with model answers to evaluate accuracy and relevance. An evaluation system based on rubrics is put into place to give marks according to important factors including conceptual understanding, accuracy, and depth of explanation. Along with optional feedback, the system creates predicted scores to ensure grading consistency and impartiality. Experimental results show that the suggested method greatly lowers evaluation time without compromising assessment reliability. The system has a lot of potential to be used in schools and online learning platforms. Future improvements include enhanced OCR accuracy, deeper contextual understanding of complex responses, and multilingual support.