AUK Digital Repository
American University Kyiv electronic data repository, also called an e-archive or centralized data repository
Communities in DSpace
Select a community to browse its collections.
Recent Submissions
DEVELOPMENT OF IMAGE CLASSIFICATION APPROACHES BASED ON DIMENSIONALITY REDUCTION METHODS
(Manuscript, 2026) Avhust, Viktor
Principal Component Analysis (PCA) is a powerful tool for reducing the dimensionality of image data while retaining critical information. This capstone project explores the application of PCA in image recognition, focusing on its use in simplifying high-dimensional image datasets to improve recognition efficiency and performance. The study focuses on two primary applications: recognizing handwritten digits (HWD) and classifying galaxies using three different types of classification. Custom software was developed to implement these image classification tasks, integrating dimensionality reduction techniques with classification algorithms to achieve high accuracy rates. The results demonstrate the effectiveness of these approaches in handling high-dimensional datasets, paving the way for using the method and implementing it in human detection software using HWD and tools to support astronomers in their research.
LEADING CROSS-CULTURAL AND REMOTE TEAMS IN HIGH-RISK ENVIRONMENTS: LESSONS FROM SAP PROJECT MANAGEMENT IN UKRAINE
(Manuscript, 2026) Khriapin, Ivan
This Capstone explores leadership in online SAP project teams composed of members from different countries who worked during the war in Ukraine. These teams operated under extremely challenging conditions, including air attacks, blackouts, stress, and remote work. While existing research highlights leadership competencies such as cultural intelligence, emotional intelligence, crisis leadership, and digital leadership, there is limited understanding of how these skills are integrated in real-life situations during a prolonged crisis. This study aims to examine how leaders behave and make decisions in such contexts.
A qualitative research design was employed. Eight SAP project managers and team leaders who collaborated with Ukrainian and international teams during the war were interviewed. The data were analyzed using reflexive thematic analysis to identify recurring patterns in leadership behavior. The findings indicate that effective leadership in this context is not defined by a single style or competency. Instead, leaders continuously balanced two dimensions: structure—including clear roles, rules, coordination, and responsibility—and care for people, encompassing empathy, emotional support, calm communication, and sensitivity to stress. These dimensions were applied simultaneously and adapted to situational demands.
Based on these insights, the study introduces the concept of Structured Empathy Leadership, defined as leading with organizational clarity while demonstrating genuine concern for team members. Rather than proposing a new theory, the study offers a practical explanation of how established leadership competencies can be integrated in high-risk environments. The findings provide valuable implications for SAP project managers and organizations working with remote and international teams in crisis or post-conflict settings.
AI IN THE BOARDROOM: DIRECTOR SENSEMAKING IN UKRAINIAN CORPORATE GOVERNANCE
(Manuscript, 2026) Skorupych, Artem
While artificial intelligence increasingly influences organizational decision-making, how corporate board directors make sense of AI for governance purposes remains underexplored, particularly in non-Western and high-stress contexts. This study investigates three questions: how boards currently engage with AI, how board composition and organizational context shape directors' perspectives, and what governance practices directors identify as necessary for responsible adoption. Using a qualitative design, the research collected data through semi-structured interviews (n=5) and written qualitative surveys (n=17) with Ukrainian board directors across banking, technology, energy, healthcare, and other sectors. Analysis followed
Gioia-inspired methodology, progressing from first-order concepts through interpretive themes to aggregate theoretical dimensions. Findings reveal directors hold a dialectical understanding of AI—simultaneously recognizing its potential to address cognitive constraints (information overload, backward focus, data fragmentation) while creating governance risks (explanation difficulties, accountability ambiguity, judgment erosion). Board composition, particularly the mix of technical and traditional expertise, systematically shapes these perspectives, while Ukrainian wartime conditions create paradoxical pressures making AI both more urgent and more risky. Directors converge on governance practices emphasizing human-in-the-loop principles, formal frameworks, transparency, and director capability-building. The study contributes to bounded rationality, upper echelons, and socio-technical systems theories while demonstrating how extreme contexts function as theoretical microscopes, revealing dynamics relevant to boards globally.
IMPROVING TASK COMPLETION PREDICTABILITY AND WORK VISIBILITY IN OUTSOURCED IT PROJECTS USING PROCESS FORMALIZATION AND AI – ASSISTED COMMUNICATION AUTOMATION
(Manuscript, 2026) Diak, Petro
Outsourced IT projects often suffer from unpredictable task completion, limited real-time visibility, and coordination inefficiencies caused by inconsistent status updates and fragmented communication. In the studied environment, Data Engineers and QA engineers update work items in Azure DevOps irregularly, follow individual reporting routines, and frequently communicate via ad hoc messages in Microsoft Teams. As a result, task boards do not reflect the actual state of work, architects and leads are overloaded with coordination tasks, and project managers lack timely information for decision-making. This reduces delivery predictability and erodes client confidence.
This Capstone addresses these issues by developing a Task Visibility and Predictability Framework that combines process formalization with AI-assisted communication automation. Using qualitative methods and semi-structured interviews, the study applies open and axial coding to identify causes of visibility breakdowns and coordination overload. Based on Agile governance and project-management practices, the framework establishes clear reporting responsibilities, standardized update routines, and structured communication rules.
The framework introduces formal update triggers, minimal documentation standards, and role-specific responsibilities. It also incorporates lightweight AI-assisted mechanisms—such as reminders, stale-task detection, and Azure DevOps–Microsoft Teams integration—to support timely updates without relying on complex models or large datasets.
Its application shows improved task visibility, fewer outdated task states, reduced need for status meetings, and lower coordination burden. The findings suggest that even simple automation, when aligned with clear processes, can significantly enhance transparency, predictability, and execution reliability in outsourced IT projects.
AN AI-DRIVEN PREDICTIVE MODEL FOR SCREENING LEGAL PROFESSIONALS IN INTERNATIONALLY ORIENTED UKRAINIAN IT COMPANIES
(Manuscript, 2026) Dzoban, Volodymyr
This applied research remedies a pressing talents issue for legal departments in Ukrainian IT companies because typical credentials are poor predictors for success in a fast-moving technology sector. The author created and tested a screening tool based on large language model and statistical modelling techniques to better evaluate candidates in an initial assessment stage. Using a dataset of 269 legal professionals who were 178 candidates for a major Ukrainian IT company and 91 professionals in the LinkedIn networking group, this study applied logistic regression analysis and narrative coding to explore predictors for success measured as acceptance of a job offer, retention for 24 months, and delivery of satisfactory performance.
Results demonstrated that a background in the information technology industry, English language skills, foreign transaction experience, interest in technology, and a business focus were strong predictors for success, while traditional credentials such as university name, prior employment with a law firm, and membership in a bar association lacked predictive power. The model has 78.4% classification accuracy with 81.2% sensitivity and 75.6% specificity in cross-validation with AUC=0.78. It was used for creating a screening tool based on GPT for candidate screening where output classifies candidate data into organized assessments with scoring points for strong traits, areas of concern, and interview recommendations. Pilot testing showed a 60% reduction in screening time while maintaining quality.
This study addresses a repeatable approach for statistical model development in the context of a particular legal staff environment and a usable screening tool for the technology sector legal recruitment. The results of this research challenge traditional forms of credentialism associated with legal recruitment and establish that culture fit factors potentially outperform legal credentials.