Bulletin of NTU "KhPI". Series: Strategic management, portfolio, program and project management

MATHEMATICAL MODEL OF VALUE CHAIN OPTIMIZATION FOR NUCLEAR SAFETY PROJECTS

Sergey Bushuyev — 2026-01-17

Nuclear safety projects are critical for ensuring the secure and sustainable operation of the global nuclear energy sector, yet they are frequently challenged by escalating costs, prolonged schedules, and complex supply chains. Traditional project management methods often fail to capture the interdependencies and high-stakes trade-offs inherent in these projects' multi-stage value chains. This paper addresses this gap by proposing a novel, integrated mathematical model for optimizing the value chain of nuclear safety projects—from design and procurement through construction and commissioning. We develop a mixed-integer linear programming (MILP) formulation that holistically integrates key decision variables, including supplier selection, logistics routing, inventory management, and activity scheduling. The model's primary objective is to minimize total lifecycle cost and project duration while treating safety, quality, and regulatory compliance as inviolable constraints. A case study based on a representative safety upgrade project is presented to validate the model. The results demonstrate the model's capability to generate optimized project plans, identify critical cost and schedule drivers, and perform robust sensitivity analysis on parameters such as resource availability and regulatory review timelines. The proposed framework provides project managers and decision-makers with a powerful, quantitative tool for strategic planning and resource allocation. By enabling a systems-level view of the project value chain, this work contributes to enhancing the economic efficiency and execution predictability of nuclear safety initiatives without compromising their fundamental safety imperative.

THE APPLICATION OF ARTIFICIAL INTELLIGENCE IN THE ANNUITY SALES BUSINESS PROCESSES OF INDEPENDENT INSURANCE AGENTS

Yevhenii Budiukov — 2026-01-17

This article presents a systematic analysis of the application of artificial intelligence (AI) in the business processes of annuity sales by independent insurance agents and proposes a process-based decomposition of the full customer interaction lifecycle. The following sequential stages are identified: lead generation; client discovery and needs assessment; product research and carrier selection; proposal presentation and consultation; application submission and underwriting; policy issuance and delivery; and post-sale engagement and retention. For each stage, a “mapping” of relevant classes of AI models and methods and their typical functions in decision support and automation is provided. A synthesis of contemporary research shows that analytical models (ML-based scoring, classification, and ensemble approaches) improve lead qualification, customer behavior forecasting, and risk assessment; generative and NLP/LLM-based solutions support work with unstructured data (call transcripts, chat transcripts, correspondence, and documents), enabling parameter extraction, summarization, and the preparation of personalized advisory materials; and agent-based approaches orchestrate workflows within CRM systems and document management. Quantitative indicators of the prevalence of model usage across process stages are presented, reflecting the uneven technological maturity of available solutions. Key implementation barriers are identified, including data fragmentation and heterogeneity, algorithmic bias, limited transparency and explainability, regulatory requirements imposed by the SEC and NAIC, and the complexity of CRM integration. Directions for future research are outlined, including the standardization of approaches, the advancement of explainable AI, and the quantitative assessment of the balance between human expertise and automation at different stages of annuity sales.

APPLICATIONS OF COMPUTATIONAL INTELLIGENCE FOR MODELING, IDENTIFICATION, OPTIMIZATION OF CONTROL INFORMATION SYSTEMS AND DECISION SUPPORT

Oleksii Kondratov — 2026-01-17

The article presents a comprehensive approach to creating intelligent control systems for dynamic objects based on advanced computational intelligence methods. In the context of processing high-dimensional, unstable, and poorly formalized data in real-time, traditional modeling methods are insufficiently effective, which is especially critical for industrial and transport systems where control errors can lead to significant losses. The research is aimed at developing adaptive closed-loop control contours that combine current state identification, prediction, and the selection of optimal control influence under uncertainty. The proposed architecture integrates modules for data collection and processing, parameter identification, optimization, and decision support. The modeling core relies on hybrid machine learning algorithms, specifically a combination of Convolutional Neural Networks and Transformer architectures, as well as metaheuristic methods. Mechanisms for online re-training with minimal weight change and the use of Universal Transformer Memory are provided for adaptation to environmental changes. Remote parameter identification is realized by analyzing multimodal data streams, including video and sensor signals, using methods of object detection, scene segmentation, and optical flow analysis. The Decision Support System has a two-level structure: a rule-based level for operational processing of typical situations and an optimization level based on multi-criteria Pareto analysis. The use of Explainable AI ensures the transparency of decisions, and model adaptation is carried out using Reinforcement Learning. The results confirm the effectiveness of the approach in industrial, transport, and urban systems and outline the prospects for further development by incorporating multi-agent and neuromorphic technologies.

TOWARDS HYBRID CLOUD INFRASTRUCTURE QUALITY ASSESSMENT MODEL

Andrii Кopp — 2026-01-17

The paper presents a hybrid model for the cloud infrastructure quality assessment, which combines subjective expert assessments with objective results of statistical analysis. The proposed model, called Hybrid Expert-derived with Entropy-based Weighted Sum Model (HEE-WSM), combines the Analytic Hierarchy Process (AHP) to determine weights based on expert assessments and the entropy-based approach to calculate weights using real data. The proposed HEE-WSM model is a novel approach that takes into account both expert judgments and cloud environment monitoring data. Eight criteria (such as availability, reliability, latency, scalability, performance efficiency, cost, security compliance, and support responsiveness) based on the international standards NIST SP 800-145 and ISO/IEC 25010 are proposed for the cloud infrastructure quality assessment. These criteria are divided into “benefit” and “cost” criteria, which is necessary to ensure normalization and proper comparison of different quality metrics. A hybrid mechanism for determining weighting coefficients allows balancing the weighting coefficients determined on the basis of AHP and the entropy approach using an adjustable coefficient that provides flexibility depending on decision-making needs. Thus, the flexibility of the proposed model is ensured by the ability to adjust the influence of subjective and objective weights of criteria. The final quality assessment is performed using the Weighted Sum Model that aggregates normalized quality metric scores for each alternative. To demonstrate the robustness of the proposed approach, ten leading cloud providers were analyzed in this study, including Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform (GCP), Alibaba Cloud, and several others. The obtained results demonstrated that the proposed model allows for effective evaluation of cloud services, with GCP receiving the highest total quality score. The proposed approach can be considered an adaptive, transparent, and useful tool for implementation in decision support systems for cloud infrastructure management. The proposed model can be applied in organizations and enterprises for the informed selection of cloud service providers. Future research includes the integration of real-time data monitoring and the application of machine learning methods for automatic adjustment of quality criteria weights.

SELECTION MODEL FOR THIRD-PARTY LIBRARIES IN IT PROJECTS

Alexander Lysenko — 2026-01-17

This study focuses on a model for selecting third-party libraries in IT projects, taking into account efforts, costs, quality, risks, technological effects, and project environment constraints. The aim of the study is to integrate these five indicators into a cohesive model that formalizes the selection process as a multi-criteria optimization problem, with clearly defined objective functions and constraints. This approach addresses a significant gap in the existing area. The study tasks include analyzing and synthesizing scientific methods for evaluating third-party libraries, establishing the relationship between evaluation criteria and project effectiveness objectives, formalizing the five key indicators into a quantitative model suitable for mathematical processing, constructing a multi-criteria optimization model for library selection that considers resource, budgetary, and quality constraints, and demonstrating the model's application through a numerical example featuring several alternative libraries. The study employs the TOPSIS method to rank alternatives based on five indicators, alongside a penalty-function mechanism to address violations of critical business constraints. The results indicate that the proposed approach allows for a unified evaluation of effort, costs, quality, risks, and technological effects of third-party libraries. It successfully normalizes diverse assessments into a decision matrix, computes aggregate measures for the overall attractiveness of alternatives, and incorporates penalties for any constraint violations. The numerical example highlights a case where the optimal choice is not the leader in individual criteria but provides an acceptable balance of integral indicators without exceeding critical constraints. Conclusions: This study provides a novel contribution by developing the first formalized multi-criteria optimization model for selecting third-party libraries. This model identifies five indicators, including technological effects, and proposes their quantitative representation. Implemented using the TOPSIS method with penalty functions, the proposed model enhances the capabilities of decision-support systems in IT projects, enabling the ranking of alternatives while accommodating resource, budget, risk, and quality constraints in complex and dynamically changing technological environments.

DEVELOPMENT AND EVALUATION OF THE EFFECTIVENESS OF AN INTELLIGENT VIDEO INTERVIEW SYSTEM FOR ASYNCHRONOUS RECRUITMENT

Vladyslav Moiseiev — 2026-01-17

The article is focused on the development and evaluation of an intellectual system for asynchronous video interviews aimed at providing objective candidate assessments and significantly reducing the time costs for recruiters. By combining advanced artificial intelligence technologies, the system analyzes candidates' responses using the GPT-4o language model from OpenAI for content and logic evaluation, while also performing emotional recognition through the DeepFace neural network, which analyzes candidates' nonverbal behavior. This combination of technologies enables not only the automation of recruitment processes but also enhances the objectivity and accuracy of hiring decisions. As a result, the implementation of this system allows companies to substantially reduce time and resource expenditures for candidate evaluation, while improving the quality of recruitment. The article discusses the main functional and non-functional requirements of the system, as well as the technical stack used for its implementation. Experimental research demonstrates that the asynchronous video interview format enables recruiters to quickly access structured analytics without the need to review all video recordings, significantly speeding up the decision-making process. Test results show high accuracy in both emotional recognition and content evaluation of candidates' responses, which increases the efficiency of recruitment and reduces the risks of subjective errors.

MODELS AND METHODS FOR CONTROLLING MULTI-GROUP CONFLICTS OF INTEREST IN A MULTI-AGENT NETWORK ENVIRONMENT OF A DISTRIBUTED SYSTEM OF INTELLIGENT AGENTS

Bohdan Skrypka — 2026-01-17

If energy resources represent an excessive value for a group of agents of a distributed system, then a logical question arises: how to effectively allocate resources in order to satisfy the needs of each agent and to distribute them sparingly so as not to overexhaust reserves? Such a problem has a non-trivial solution. Our work is based on such models that were studied, inspired by the natural form of existence of group biological beings. The object is intelligent distributed systems. The subject is the process of self-organization of intelligent agents into a holistic distributed system that has common interests and the study of effective strategies for resolving emerging conflict situations between agents. Relevance of the work. Without energy resources, the existence of a particular biological species on the planet is impossible, and therefore the balance and distribution of energy is of key importance. Electricity is currently a scarce resource – the question of the priority of distribution between different groups of agents of a distributed network is a task not only for energy engineers, but also for mathematicians and economists. Water, human resources, mechanical means of shared use, etc. can act as a controversial or conflicting resource for several agents that need to be resolved. The goal of this work is to analyze and synthesize current knowledge on the topic of research into new approaches-strategies for organizing intelligent agents of distributed systems and to form ideas, hypotheses, observations and experimentally obtained data, methods and approaches for resolving conflict situations between agents into text-graphic technology for further formation of a mathematical model. The main idea consists in forming a new methodology-strategy for resolving multi-group conflict situations in distributed systems, which will potentially allow for the release and redirection of additional resources to achieve the ultimate goal. Methods used. Scientific experiment, analysis and synthesis, methods of comparison and analogy, modeling and prototyping method, abstraction and concretization, observations. Results obtained: scientific materials in the area of this research were analyzed and synthesized into local knowledge; an analysis of tactical and strategic methods for resolving conflict situations in the context of multi-agent systems was conducted; new knowledge was formed in the form of text-graphic technology; possible problem statements were considered at the verbal level for resolving conflict situations and resolving intergroup conflict using the example of coordination of intelligent agents of a wolf pack. Further developed. The concept and the very idea of resolving conflict situations based on the mathematical apparatus in the problems of modeling and optimizing the use of resource indicators in the context of multi-group interaction of agents of a distributed intelligent network using the example of organizing a wolf pack; the concept of finding an effective solution to the N-dimensional prisoner's dilemma. Scientific novelty. In this work, it is proposed to solve the N-dimensional prisoner's dilemma using the algorithmic method of controlled multi-group conflict based on the metaheuristic algorithm of the swarm intelligence of a pack of gray wolves (GWO). The algorithm model of a pack of gray wolves has become possible to transfer to the N-dimensional formulation of the prisoners' dilemma problem due to the fact that the ability of an individual wolf agent to act as an independent individual and influence the overall outcome of cooperation of wolves in one pack is the prototype of the game model, which is represented by the "N-dimensional prisoners dilemma". Thus, each wolf agent is able to influence the level of decision-making efficiency of the leader of his pack, who represents the player of the "N-dimensional prisoners' dilemma", who tries to maximize his (the pack's) gain on the solution search plane. Practical significance. The proposed methods, models and techniques can be used in applied problems of economic calculation of enterprise efficiency, in problems of mathematical modeling of finding balance and effective allocation of resources, in problems of searching and identifying and surrounding dynamic goals in N-dimensional environments, achieving an effective fitness function indicator even on multi-extreme optimization functions, in problems of economical allocation and consumption of resources. Conclusions. The tasks specified in the objective were fulfilled and a text-graphic technology was created for the further formation of a mathematical model for solving the problem of polygroup conflict of a multiagent system.

SENSOR MODALITIES FOR RELATIVE LOCALIZATION OF COOPERATIVE UAV SWARMS IN GNSS-DENIED ENVIRONMENTS

Vitalii Tymofieiev — 2026-01-17

The deployment of cooperative swarms of unmanned aerial vehicles in environments without access to global satellite navigation systems represents one of the key challenges in modern autonomous multi-agent systems. Under such conditions, relative localization among agents must provide sufficient accuracy and consistency while operating under strict constraints on size, weight, and power consumption, which are characteristic of mass-produced micro-scale platforms. This paper presents a systematic review of sensor modalities used for relative localization of cooperative swarms in navigation-denied environments, with a particular focus on their computational properties, observability, and robustness to environmental factors. The study analyzes and classifies three principal classes of sensor modalities: radio-frequency distance measurement based on ultra-wideband communication, passive visual methods including visual–inertial odometry and deep learning–based approaches, and systems employing active optical markers. The analysis demonstrates that radio-frequency ranging methods offer low computational cost but suffer from fundamental observability limitations due to the lack of angular information. Passive visual approaches are capable of achieving high accuracy and global consistency; however, they require substantial computational resources, which restricts their practical applicability on platforms with severe hardware constraints. A critical evaluation indicates that active optical marker systems, when combined with distributed state graph optimization methods, constitute a practically viable compromise between localization accuracy, computational efficiency, and robustness under degraded environmental conditions. Particular attention is given to distributed estimation architectures that enable scalability and consistency without centralized processing. The paper concludes by outlining directions for future research aimed at developing hybrid frameworks capable of dynamically switching between full relative pose estimation and direction-only tracking in challenging operational environments.

THE DISPATCHER: BRIDGING THE PROBABILISTIC GAP IN AUTOMATED DECISION MODELING

Olga Cherednichenko — 2026-01-17

In the contemporary landscape of Software Engineering and Business Process Management (BPM), the integration of generative artificial intelligence has precipitated a paradigm shift from manual, deterministic specification to automated, probabilistic generation. While offering scalability, this transition introduces a fundamental volatility known as the "Probabilistic Gap"—the chasm between the fluid, high-variance output of Large Language Models (LLMs) and the strict, zero-tolerance syntactic requirements of execution engines like DMN (Decision Model and Notation). This paper addresses the "Struc-Bench Paradox," highlighting the limitations of transformer architectures in generating complex structured data without rigid orchestration. The study formally defines and implements the "Dispatcher," a pivotal control plane component designed to function as an intelligent resource arbiter and quality gatekeeper within a neuro-symbolic architecture. The theoretical framework shifts the economic focus from Baumol’s Cost Disease, which addresses production speed, to Boehm’s Law of Software Economics, which emphasizes the exponential cost of defects propagated to production. To operationalize this, the Dispatcher represents a discrete deterministic process modeled using Cost-Colored Petri Nets rather than Finite State Machines (FSMs). The Petri Net formalism allows for precise modeling of concurrency, state accumulation, and the strict enforcement of "Retry Budgets," thereby mathematically guaranteeing system termination and preventing infinite loops of costly regeneration. The architectural implementation utilizes a "Test-First" generation philosophy: the system first synthesizes validation criteria (JSON test cases) utilizing Schema Injection and RAG, and subsequently grounds the generation of DMN logic (XML) in these pre-validated scenarios. Experimental analysis was conducted using a controlled set of 200 generation cycles to evaluate two distinct error-recovery strategies: Strategy A (Independent regeneration of DMN tables only) and Strategy B (Joint/Dynamic regeneration of both DMN and Test Cases). Quantitative results demonstrate that Strategy B is economically superior, achieving a 6.06% reduction in total cost and an 8.44% reduction in token consumption compared to the independent patching approach. The findings indicate that simultaneous regeneration empowers the LLM to resolve semantic incoherence and hallucinations more effectively than iterative repairs, prioritizing logical consistency over partial code retention. The study concludes that the Dispatcher effectively bridges the neuro-symbolic divide by transforming validation from a post-production manual review into a pre-production automated cycle. By enforcing a "Stop-Loss" mechanism driven by economic constraints, the framework minimizes the Total Cost of Ownership and serves as a critical "Trust Proxy," mitigating automation bias and ensuring that AI-generated artifacts meet the rigorous reliability standards required for enterprise deployment.