CogniGrade: An AI-Powered System for Automated Grading of Handwritten Answers

The process of manually evaluating handwritten an-swer sheets is time-consuming, error-prone, and often influenced by subjective judgment. With advancements in deep learning and computer vision, automated grading systems are emerg-ing as efficient alternatives to traditional evaluation methods. CogniGrade presents an AI-driven approach for the automated assessment of handwritten responses, combining optical character recognition (OCR), diagram interpretation, and semantic anal-ysis. The system first digitizes handwritten text using advanced OCR models such as CRAFT for text detection and TrOCR for text recognition. For diagram-based questions, a YOLOv5-based object detection model identifies and interprets various flowchart components. Both the recognized textual and visual data are semantically analyzed through a language model to generate an accurate and consistent evaluation score. The proposed system aims to minimize human effort, enhance grading consistency, and significantly reduce evaluation time, thereby improving the efficiency and fairness of academic assessments.

Paper

Full text

PDF

CogniGrade: An AI-Powered System for Automated Grading of Handwritten Answers

Semantic Scholar · 2026

Abstract

The process of manually evaluating handwritten an-swer sheets is time-consuming, error-prone, and often influenced by subjective judgment. With advancements in deep learning and computer vision, automated grading systems are emerg-ing as efficient alternatives to traditional evaluation methods. CogniGrade presents an AI-driven approach for the automated assessment of handwritten responses, combining optical character recognition (OCR), diagram interpretation, and semantic anal-ysis. The system first digitizes handwritten text using advanced OCR models such as CRAFT for text detection and TrOCR for text recognition. For diagram-based questions, a YOLOv5-based object detection model identifies and interprets various flowchart components. Both the recognized textual and visual data are semantically analyzed through a language model to generate an accurate and consistent evaluation score. The proposed system aims to minimize human effort, enhance grading consistency, and significantly reduce evaluation time, thereby improving the efficiency and fairness of academic assessments.

Similar papers

© 2026 NYSGPT2525 LLC