Design and Development of a Human-Evaluation Platform for Large Language Model-Generated Datasets

This study aims to design and develop a human evaluation platform for assessing educational question-answer (QA) datasets generated by closed-source large language models (LLMs). As the use of generative AI expands in educational settings, there is a growing need for structured review systems capable of verifying the reliability and curricular alignment of high-quality datasets. The proposed platform enables experts to evaluate QA datasets by uploading data, ranking multiple responses, and providing qualitative feedback. To support this process, a web-based system was implemented, incorporating features such as metadata management, expert account creation and assignment, and rank-based evaluation. By introducing a practical tool that supplements human involvement in dataset validation, this study lays the groundwork for the evaluation and refinement of educational language models.