DEVELOPMENT OF EDUCATIONAL ECOSYSTEMS IN THE CONTEXT OF AI MODEL SPECIALIZATION

Abstract

The systemic evolution of Artificial Intelligence (AI) tools in higher education is currently shifting from general-purpose generative models toward specialized agentic architectures. In the context of modern pedagogical science, this shift is driven by the limitations of mainstream AI solutions, which often lack accuracy in specific scientific domains and exhibit complexity when applied to comprehensive educational scenarios. This necessitates the implementation of specialized environments capable of ensuring high factual reliability while optimizing the cognitive resources of both students and educators. According to the updated European Digital Competence Framework (DigComp 3.0, 2025), developing the capacity for critical algorithm management and the use of verified knowledge bases has become a strategic priority for professional teacher training. The study aims to substantiate a multimodal agentic ecosystem as a methodological framework for inquiry-based learning that enhances academic integrity and professional competence in informatics teacher training. This model defines the AI agent not merely as a chatbot but as a functional instrument for intellectual support, bridging the gap between advanced architectural solutions and practical pedagogical strategies. The methodology combines a systemic approach to AI architecture analysis, a synthesis of cognitive load theory, and empirical verification via a case study involving the NotebookLM platform. This involved an analysis of how agentic workflows facilitate the transition from linear information retrieval to iterative research cycles. Furthermore, a comparative assessment was conducted to evaluate the efficiency of the proposed framework in automating routine data structuring. The results confirm that the proposed model is a valid framework, showing strong alignment with the requirements for multimodal knowledge integration across text, code, and visuals. A central finding is the effectiveness of RAG-architecture in grounding AI responses in verified professional corpora, such as national educational guidelines. Testing within a «Prompt Engineering» course demonstrated that specialized agentic systems significantly minimize factual errors, allowing students to shift their cognitive focus from data verification to high-level synthesis and critical evaluation of educational content. The research concludes that modern informatics teacher training must integrate multimodal agentic systems to prepare specialists capable of managing dynamic knowledge bases. Ultimately, this transforms the teacher’s role from a knowledge transmitter to a designer of the learning environment, coordinating the «teacher – AI-agent – student» interaction within a unified digital ecosystem.