Integration of a neural network assistant on programming into a digital educational platform

The article presents the results of the development and implementation of instrumental student support based on generative large language models (LLM) within a digital educational platform (DEP). The problems of this study are caused by the lack of operational methodological feedback in the process of mass programming training, especially at the initial stages of developing practical skills. The proposed solution is to create and integrate a neural network module that aggregates information about a learning task (condition, name, hidden tests, reference solution, etc.) to generate personalized pedagogical prompts aimed at step-by-step comprehension of mistakes made and their independent correction by students without directly giving the correct solution. The article describes in detail the service architecture, the principles of its integration into the task verification system, as well as mechanisms for validating model responses and protecting them from misuse or unfair use. The empirical base of the study includes the results of testing in the framework of intensive specialized shifts (mini-courses) “Kurchatov Codes” conducted for high school students, using quantitative methods of analyzing educational and behavioral data. It is shown that the use of a neural network assistant helps to increase student engagement, reduce cognitive and psychological barriers in finding and correcting mistakes, and leads to a statistically significant reduction in the routine burden on teaching staff while maintaining the pedagogical validity of training.