Fair grade AI: An Intelligent System For Transparent And Unbiased Exam Evaluation

Abstract

 Fair and consistent evaluation of descriptive and handwritten examination answers remains a significant challenge in modern education due to subjectivity, evaluator bias, and time constraints. This paper presents FairGrade AI, an intelligent exam evaluation system designed to automate and standardize the grading process while maintaining transparency and interpretability. The proposed system integrates Optical Character Recognition (OCR) to digitize handwritten responses, followed by a Retrieval-Augmented Generation (RAG) framework that retrieves relevant evaluation criteria and domain-specific reference material. A large language model (LLM), guided through structured prompt engineering, evaluates student answers strictly based on the retrieved context to ensure criteria-aligned scoring. The system generates not only marks but also detailed explanations, identified gaps, and actionable feedback, thereby enhancing both assessment quality and learning outcomes. The architecture is implemented using a modular full-stack approach with React for the frontend, FastAPI for the backend, MongoDB for data management, and a vector database for semantic retrieval, with scalable deployment supported by cloud services. Experimental evaluation demonstrates improved consistency and reduced grading time compared to traditional manual methods, while maintaining close alignment with human assessment standards.The proposed framework demonstrates how the integration of OCR, retrieval-augmented generation, and large language models can enable a more transparent, scalable, and consistent approach to automated exam evaluation in educational settings.