Information technologies have become an essential part of the educational process and are evolving so rapidly that keeping track of newly available tools for organizing training sessions is quite challenging. One of such innovative tools that can be integrated into education is chatbots powered by large language models. Large Language Models (LLMs) are advanced artificial intelligence algorithms trained on vast amounts of data for natural language processing [1]. These models can generate text, analyze images, create new content, and propose new ideas.

Using LLMs in education can be an effective way to support both students and educators. However, it is important to consider that the generated responses may not always be entirely accurate [2], requiring critical evaluation and analysis. This, in turn, fosters the development of critical thinking skills.

To facilitate interaction between students and educators with LLMs, a chatbot-based interface was implemented, allowing real-time requests and instant responses within a predefined “thinking paradigm.” The concept of the student question-answering service was to develop a chatbot that simulates various historical figures, enabling real-time conversations. The bot generates responses based on the contextual knowledge available for a given historical character.

As part of the ongoing research on the creation of a historical tutor chatbots, two prominent figures were chosen. he first is Konstantin Ushinsky, a Russian educator, writer, public figure, and founder of scientific pedagogy in Russia. The second is Lev Vygotsky, known for his concepts of the zone of proximal development, cultural-historical theory, and a learning approach based on social interaction. A fully trained LLM, developed using the works of these scholars, can accurately respond to questions in alignment with their respective educational theories, acting as a knowledgeable successor. The model can provide quotations, draw comparisons, and formulate hypotheses.

A key feature of this service is its ability to handle routine educational tasks in various roles, such as expert, examiner, course developer, or webinar presenter. It can also generate materials to overcome the "blank page" challenge by generating test questions, discussion prompts, and case studies. The effectiveness of this approach depends on well-defined professional tasks and clear input parameters. The developed chatbots were named “Ph.D. Ushinsky” and “Ph.D. Vygotsky,” respectively.

The Retrieval-Augmented Generation (RAG) method was applied to fine-tune and adjust the focus of a large language model [3]. This method relies on retrieving and extracting relevant information to enhance the user's query. In this process, after the prompt is generated using RAG, it is expanded with additional context before being fed into the LLM. The model then produces a response based on the enriched input. The Telegram messenger was chosen as the chat platform, and a commercially available LLM was selected, with the flexibility to be replaced by any equivalent model, including a locally hosted LLM. Using the RAG approach, a database of text excerpts and their vector representations was built to facilitate the retrieval of relevant content from historical texts. LanceDB (database) was used to store these vector embeddings. When selecting materials from the works of historical figures serving as tutors, the content was organized by topic. Key areas such as upbringing, education, development, and culture were identified. Important quotes and summaries relevant to chatbot responses were extracted. Additionally, practical examples from historical texts were highlighted to create interactive dialogues and realistic scenarios for the chatbot. To evaluate the performance of the LLM-based chatbots, a panel of experts conducted a comparative analysis of the initial model and the refined versions, assessing them across 14 criteria, including “Accuracy of concept definitions,” “Completeness of responses,” “Consistency with the historical figure’s ideas” and others. The “Ph.D. Vygotsky” chatbot provided precise and concise definitions aligned with original sources, ensuring clarity without unnecessary detail. In contrast, the “Ph.D. Ushinsky” chatbot occasionally mixed up concepts, delivering generally correct but incomplete responses. Notably, the “Ph.D. Vygotsky” model demonstrated the most accurate and comprehensive understanding of the “zone of proximal development” concept, offering well-structured answers that fully adhered to Lev Vygotsky’s terminology. Both chatbots performed well in terms of “relevance,” “practical applicability,” and “validity,” demonstrating their ability to translate theoretical knowledge into practice. “Ph.D. Ushinsky" and “Ph.D. Vygotsky” both exhibited strong language proficiency, reflected in their clarity, appropriate terminology, and contextual awareness. However, “Ph.D. Vygotsky” showed greater theoretical depth and analytical ability, while “Ph.D. Ushinsky” excelled in the practical application of theory and exhibited a higher level of creativity.

Chatbots trained on classical pedagogical texts represent a new phase in the evolution of educational technology. Specifically, the chatbot developed based on Konstantin Ushinsky’s works provides valuable opportunities for structuring the educational process in pedagogical universities. Its primary advantage is its ability to integrate the principles of classical pedagogy into modern education while adapting to the learning styles and needs of today's students. A chatbot can serve as a mentor for students. It can explain topics or concepts in simple terms, just as Konstantin Ushinsky would, and offer guidance on organizing the educational process, selecting teaching methods, and addressing other pedagogical issues. Students can interact with “Ph.D. Ushinsky” by discussing the teacher’s role in shaping a student’s personality or exploring the use of game-based teaching methods, receiving feedback from Ushinsky’s perspective. With the assistance of these chatbots, students can evaluate their knowledge. The chatbot can generate quizzes and provide personalized feedback on incorrect answers. Additionally, it can help students identify key topics, create educational activities such as crossword puzzles, and support speech preparation by suggesting themes and titles.

1. P. Kumar, “Large language models (LLMs): survey, technical frameworks, and future challenges,” Artif Intell Rev, vol. 57, no. 10, p. 260, Aug. 2024, doi: 10.1007/s10462-024-10888-y. 2. S. S. Gill et al., “Transformative effects of ChatGPT on modern education: Emerging Era of AI Chatbots,” Internet of Things and Cyber-Physical Systems, vol. 4, pp. 19–23, 2024, doi: https://doi.org/10.1016/j.iotcps.2023.06.002. 3. M. Arslan, H. Ghanem, S. Munawar, and C. Cruz, “A Survey on RAG with LLMs,” Procedia Computer Science, vol. 246, pp. 3781–3790, Jan. 2024, doi: 10.1016/j.procs.2024.09.178.