Along with the social interest in generative artificial intelligence brought about by ChatGPT, attempts are steadily being made to apply it to education. Unlike conventional rule-based chatbots, ChatGPT-based generative chatbots are meaningful in that they can create sentences by newly constructing meanings based on ultra-large AI and present various cases using a large amount of data. This will help students view scientific concepts and cases in various directions and organize knowledge in a form that suits them. In this study, an educational chatbot was developed in which the GPT-3.5 model was Fine-tuned according to the level of sixth graders in elementary school in Korea and the contents of the science curriculum. The GPT-3.5-turbo-16k-0613 model provided by OpenAI was utilized and fine-tuned to suit the purpose. In this process, it was found that not only sufficient datasets should be secured in advance, but the more detailed and detailed the System settings were, the more selected answers were generated according to the developer's intention.