AnswerKey Validation Intelligence: AI- Assisted Exam Evaluation Framework

Abstract

Manual assessment of descriptive examination answers requires substantial faculty time and can produce variation when different evaluators interpret the same response. This paper presents AnswerKey Validation Intelligence, a web-based system designed to assist—not replace—faculty in processing and evaluating descriptive answer sheets. The proposed workflow combines optical character recognition (OCR), natural language processing (NLP), semantic comparison with faculty-provided model answers, keyword-based evidence, and human verification. Students submit answer sheets as images or PDFs; the system extracts text and organizes responses for question-level comparison. Preliminary marks are generated from configurable evaluation criteria and presented to faculty for correction and approval before publication. The platform also supports examination management, role-based access, record storage, and student result viewing. The design draws on prior work in automatic short-answer grading, contextual language representations, sentence embeddings, and transformer-based OCR. The project report describes the application workflow and its main modules, but does not provide a sufficiently documented benchmark dataset or numerical grading-accuracy measurements. Accordingly, this paper focuses on the system design, implementation workflow, and a reproducible evaluation plan rather than claiming unmeasured performance gains. A key design principle is that OCR errors and semantic scores must remain reviewable so that final marks are approved by a faculty member.

Introduction

The paper proposes AnswerKey Validation Intelligence, an AI-assisted automatic descriptive answer evaluation system designed to reduce the time and effort required by teachers to manually evaluate students' answer sheets.

Traditional descriptive answer evaluation is time-consuming because teachers must read every answer, compare it with the model answer, identify important concepts and award marks based on correctness and partial credit.

The proposed system uses Optical Character Recognition (OCR), Natural Language Processing (NLP) and semantic similarity techniques to extract and evaluate students' answers. It generates preliminary marks, which faculty members can review, modify and approve before publishing the final results.

1. Introduction and Problem Statement

Manual evaluation of descriptive answers creates several challenges:

• Requires significant time and effort from teachers.

• Becomes difficult when there are many students.

• Exact keyword matching cannot identify all correct answers.

• Students may express the same concept using different words.

• Handwritten answers are difficult to process because of handwriting variations.

• OCR errors can affect the accuracy of automatic evaluation.

The proposed system addresses these challenges by combining OCR, semantic matching, keyword detection and faculty verification.

2. Objectives

The main objectives of the project are:

• Develop a centralized platform for examination management.

• Allow faculty members to create examinations and model answers.

• Allow students to upload handwritten answer sheets.

• Extract text automatically using OCR.

• Compare student answers with faculty-provided model answers.

• Use semantic similarity and keyword matching for evaluation.

• Generate preliminary marks automatically.

• Allow faculty members to verify and modify marks.

• Maintain secure examination records and publish approved results.

3. Related Work

The proposed system is based on existing research in automatic answer grading, semantic similarity and OCR technologies.

Important research and technologies include:

• BERT: Understands the context of words and sentences.

• Sentence-BERT: Generates sentence embeddings for semantic similarity comparison.

• Tesseract OCR: Converts document images into machine-readable text.

• TrOCR: Uses Transformer architecture for printed and handwritten text recognition.

• Automatic Short Answer Grading: Provides techniques for evaluating descriptive answers using linguistic and machine learning methods.

These technologies provide the foundation for the proposed system, but their published performance does not guarantee the accuracy of this particular implementation.

4. Proposed System

The system provides three types of users:

1. Administrator: Manages users, access permissions and system records.

2. Faculty: Creates examinations, uploads model answers, defines keywords, reviews marks and publishes results.

3. Student: Uploads answer sheets and views results after faculty approval.

The complete workflow consists of six major stages:

1. Faculty creates an examination and provides model answers.

2. Students upload answer sheets in image or PDF format.

3. OCR extracts handwritten or printed text.

4. Extracted answers are mapped to their respective questions.

5. AI compares student responses with model answers and generates preliminary marks.

6. Faculty reviews and approves the marks before publishing results.

Conclusion

This paper presented AnswerKey Validation Intelligence, a faculty-supervised system for supporting descriptive answer evaluation. The workflow combines OCR-based text extraction, question-wise comparison with model answers, semantic similarity, keyword evidence, faculty verification, and structured result management. The system is intended to reduce repetitive handling of answer sheets while retaining human control over final academic decisions. Its practical value depends on the quality of OCR output, the appropriateness of the marking rubric, and the reliability of the semantic scoring method. The current project documentation establishes the system concept and workflow but does not provide sufficient quantitative evidence to support claims of improved grading accuracy or measured time savings. The next step is to conduct a controlled evaluation with anonymized answer scripts and faculty-assigned reference marks, report OCR and grading metrics, and analyze the cases in which the system disagrees with evaluators. Further development may include question-specific rubrics, better response segmentation, support for additional languages and notation, calibrated partial-credit rules, audit logs, and feedback to help faculty identify uncertain predictions. The system should be treated as an assessment assistant rather than an autonomous authority on student performance.

References

[1] S. Burrows, I. Gurevych, and B. Stein, “The Eras and Trends of Automatic Short Answer Grading,” International Journal of Artificial Intelligence in Education, vol. 25, pp. 60–117, 2015, doi: 10.1007/s40593-014-0026-8. [2] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Lnguage Understanding,” in Proceedings of NAACL-HLT, vol. 1, pp. 4171–4186, 2019, doi: 10.18653/v1/N19-1423. [3] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks,” in Proceedings of EMNLP-IJCNLP, pp. 3982–3992, 2019, doi: 10.18653/v1/D19-1410. [4] R. Smith, “An Overview of the Tesseract OCR Engine,” in Proceedings of the Ninth International Conference on Document Analysis and Recognition (ICDAR), pp. 629–633, 2007, doi: 10.1109/ICDAR.2007.4376991. [5] M. Li et al., “TrOCR: Transformer-Based Optical Character Recognition with Pre-trained Models,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 11, pp. 13094–13102, 2023, doi: 10.1609/aaai.v37i11.26538.

Copyright

Copyright © 2026 Mrs. Dhana Lakshmi R, Vishal Suriya K R, Thirumurugan S, Thilip T. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.