The article substantiates and systematizes approaches to the use of artificial intelligence (AI) technologies in higher professional education as a tool for information and analytical support of individual training, coaching, management activities and psychological counseling. The purpose of the work is to present ways to use AI tools (generative models, intelligent assistants, recommendation systems and intelligent document processing tools) in the educational process and university management while maintaining human responsibility and the requirements of academic ethics. It is shown that the key effect of AI in higher education is the transition from “average” learning to personalized trajectories supported by educational and subject analytics, as well as the formation of new mentoring (coaching) practices based on data. Solutions (analytical recommendation systems; intelligent educational environments; management support systems) are proposed and typical scenarios are described: an AI tutor, a coaching scenario, an analytical dashboard, automation of teacher training, support for psychological counseling (query screening, dialogue simulators). The risks (hallucinations, data bias) and the need for verification by a source specialist are highlighted separately. The practical significance of the work lies in the formation of applicable approaches to the introduction of AI in higher education, focused on improving the quality of education and the manageability of educational processes. The following theoretical methods were used in the development: (1) analysis and synthesis, a systematic approach, comparative analysis, classification, generalization of psychological-pedagogical, organizational and managerial experience in the implementation of AI; (2) information and analytical methods — content analysis of scientific publications and practices of AI application in higher education; analysis of typical application cases (generative assistants, chatbots); (3) methodological modeling — designing a methodological model for the use of AI (levels: training / coaching / management / consulting), description of functions, inputs/outputs and constraints (including requirements for data, validation and ethics); (4) identification of risks and measures to minimize them, verification by sources, regulations for disclosure of the use of AI, data protection. 

The digital transformation of education has led to exponential growth in learning activity data, creating technological prerequisites for the implementation of intelligent recommender systems (IRS). However, commercial recommendation algorithms optimised for maximising engagement and profit cannot be directly transferred to the educational environment, where the main objective function is to improve the quality of knowledge acquisition and reduce academic failure. Aim. To develop a conceptual model of an intelligent recommender system with explainable artificial intelligence mechanisms and built-in ethical control tools. Methods. The research is based on a systematic review of peer-reviewed for 2021–2025 indexed in Scopus, Web of Science, and RSCI. Methods of comparative analysis of recommendation algorithms, architectural modelling, and experimental validation on open educational datasets. Results. A four‑level IRS architecture is proposed, comprising a data collection layer, a processing and vectorisation layer, an algorithmic core based on a hybrid model, and an interaction layer with a recommendation explanation module. Experiments have shown that the proposed hybrid model outperforms classical collaborative filtering by 13% in precision and solves the cold‑start problem through semantic analysis of new courses. The developed explanation module, based on the LIME method, visualises the factors influencing each recommendation, thereby increasing trust among teachers and students. Conclusions. The results are consistent with the findings of leading international researchers on the advantages of hybrid graph models over purely collaborative approaches. Unlike known solutions, the proposed model includes a built‑in fairness checking block and is designed for integration with Russian learning management systems. 

In the context of the digitalization of the banking sector and the increasing requirements for the speed and security of cash handling, traditional methods of cash-in-transit (CIT) operations no longer provide the necessary level of efficiency. An analysis of existing solutions shows that modern CIT services face fragmented technological infrastructures, insufficient integration between mobile and web systems, and a lack of intelligent decision-support tools. The gap in research literature lies in the absence of comprehensive architectures that unify mobile applications, web-based monitoring services, and artificial intelligence modules into a single ecosystem for managing CIT processes. The purpose of the study is to analyze the possibilities for automating CIT processes using mobile and web technologies, as well as intelligent components aimed at improving the efficiency, transparency, and security of cash-in-transit operations through a mobile automated CIT system. The methodological foundation includes systems analysis of CIT processes, business process modeling (BPMN), architectural design, software engineering methods, and the application of machine learning algorithms for route optimization, delay prediction, and anomaly detection. The results demonstrate that the use of a mobile application in combination with a monitoring center web service and AI agents enables continuous data exchange, improves planning accuracy, automates briefing and documentation procedures, and reduces the impact of human factors. The conclusions confirm that the implementation of the Mobile Automated Cash-in-Transit System (MACTS) ensures a significant increase in the efficiency of CIT processes, enhances security, and establishes a foundation for the further development of intelligent solutions in logistics and banking automation. The prospects of the research include expanding the functionality of AI agents and integrating the system with corporate security and financial monitoring platforms

The article examines the use of artificial neural networks for managing information security risks in electronic document management systems. The paper considers the transition from traditional protection tools based mainly on signature rules and expert procedures to adaptive models for security event analysis. Particular attention is paid to the specifics of electronic document management as a digital environment that simultaneously processes legally significant documents, personal data, business correspondence and technological logs. The study shows that the rapid growth of electronic document flows, the distributed nature of organizational processes and the increasing complexity of network infrastructure require a revision of approaches to monitoring and prioritizing threats. As a promising solution, the article proposes a neural network-based risk-oriented security framework that includes a sensor layer for event collection, an intelligent TrafficLLM correlation module and a continuous retraining mechanism based on EA-PEFT parameter-efficient adaptation. This architecture makes it possible to take data drift into account, detect non-standard patterns of user and network behavior, and generate risk assessments without significantly increasing the operational workload. The scientific novelty of the study lies in the development of an integrated model for applying neural network technologies to EDMS information security risk management: from the classification of organizational, administrative, subjective and technological risks to the description of model integration into electronic document management infrastructure. The practical significance consists in the possibility of applying the proposed approach in organizations with distributed structures, intensive circulation of legally significant electronic documents and strict requirements for business process continuity.

In The context of the digital transformation of Russian education, regulated by national projects, the adoption of intelligent chatbots based on large language models (LLMs) is becoming particularly relevant. The study aims to systematize the experience of using multimodal chatbots in education and assess the prospects for their adaptation to the Russian national messenger MAX (VK). Based on an analysis of scientific publications from 2018 to 2025, key areas for the effective use of bots have been identified: personalized learning, automation of routine processes, instant 24/7 feedback, gamification, etc. Optimal architectural solutions have been determined, such as hybrid systems (YaLM + RAG) and multimodal orchestration (text/graphics/audio). Key risks and technical, pedagogical, and ethical limitations have been identified. The prospects for development are outlined, which include: a full transition to domestic LLMs (GigaChat, Saiga), the integration of virtual and augmented reality (VR/AR) into educational chatbots, the creation of a federal platform for educational bots, and the development of discussion forums and a teacher training system for adapting to digital tools. This article will be useful for researchers and practitioners interested in the implementation of AI technologies in education and lays the groundwork for developing new educational solutions at the intersection of pedagogy and artificial intelligence.

Modern social networks are characterized by a complex non-random topology; however, existing research provides only a fragmented description of their structural properties within the Russian segment, leaving a gap in the validation of classical graph models. The relevance is driven by the need to develop precise mathematical analysis methods for managing information flows and countering threats in the digital environment. The aim of the work is the modeling and structural analysis of the “VK” social network using graph theory to identify key metrics of clustering, the “small-world” effect, and scalability. The study is based on the analysis of an anonymized data sample from the “VK” social network comprising over 1 million users. An undirected graph of friendship ties was constructed. Algorithms were applied to calculate the clustering coefficient, node degree distribution, average shortest path length, and modularity using the Louvain method. It was established that the “VK” network exhibits properties of a scale-free network with a power-law degree distribution and the presence of hubs, a high clustering coefficient (≈0.52), and a short average path length (<5 steps), which corresponds to the Barabási–Albert model and the “small-world” effect. Stable thematic communities with high structural closure were identified. The obtained results correlate with the findings of Milgram, Watts-Strogatz, and Barabási-Albert, confirming the universality of these models for Russian platforms. Research prospects involve the study of dynamic information diffusion processes and the application of graph neural networks for link prediction.

This article proposes a solution to the problem of socioeconomic synthesis of big data using informatics. The dimensionality of economic processes converges with the dimensionality of social processes, which is a novel feature of the research object of leading data. It reframes the subject of research as the intersection of econometrics and sociology in informatics, transforming methods of socioeconomic analysis, and searching, processing, and analytical presentation of big data. The article examines the socioeconomic aspects of updating leading data and informatics as a platform for transforming econometric methods for synthesizing big data. The conclusions advance the idea of big data technology to the forefront of economics and management, and presents fundamental methodological, organizational, and technical measures for introducing leading data technology into new, higher economic paradigms.

Currently, theoretical and practical research pays much attention to merger and acquisition deals, but it is of great importance to study behavior factor and irrational behavior buying company too. The anchoring effect enables estimating this type of deviations on the quantitative level. The relevance of the study is provided by the necessity to take into account investor cognitive bias to explain logic and to evaluate efficiency of the national stock market. Academic novelty is an empirical verification of the hypothesis applied to Russian capital market with the use of quantitative methods. The aim of the paper is to test the hypothesis concerning the influence of anchoring effect on acquiring company’s stock returns on Russian M&A market and to perform quantitative estimation of the strength of this influence by regression modeling. The outcome of the study is the practical recommendations on taking into account the behavior bias designed for investors. The research was carried out on the basis of data on domestic acquiring companies by means of correlation and regression analysis. The methodology comprised modeling of expected and abnormal stock returns relative to merger and acquisition deals date announcement to reveal statistically significant relationship. The study empirically proves a significant anchoring effect on quotation dynamics. The regression equations have been obtained, statistical significance of factors has been determined that is a quantitative measure of the given behavior bias on Russian market. The obtained results concord with the results of the studies performed for foreign stock markets, proving universality of the effect. Practical value of the research is the possibility to implement this approach for comparative analysis of efficiency of various capital markets and for development of investment strategies addressing the consequences of behavior prejudice.