Leadership development has become an essential component of software engineering education, particularly within capstone projects that simulate real-world collaborative environments. This study investigates how undergraduate students develop leadership competencies through participation in capstone software projects by employing a qualitative analysis of secondary data sources, including peer-reviewed literature, case studies, and documented student experiences. Guided by experiential learning theory, transformational leadership theory, and social constructivism, the research explores the processes through which leadership emerges, evolves, and is enacted in student teams. The findings reveal that leadership development is dynamic and context-dependent, shaped by communication practices, team interactions, decision-making responsibilities, and mentorship structures. While capstone projects provide rich opportunities for leadership growth, outcomes are often uneven due to variations in team composition, institutional support, and assessment practices. The study highlights the importance of integrating structured leadership training, reflective learning, and aligned evaluation mechanisms within software engineering curricula. By offering a comprehensive synthesis of existing knowledge, this research contributes to a deeper understanding of leadership formation and provides practical implications for enhancing educational practices in computing disciplines. Keywords: leadership development, capstone projects, software engineering education, experiential learning, transformational leadership, student teamwork, qualitative analysis

Medical visual question answering (Med-VQA) combines medical image understanding and natural-language processing to answer clinically relevant questions about radiographs, computed tomography, magnetic resonance imaging, pathology, and other visual data. This study evaluates existing Med-VQA methods from secondary sources and proposes a telemedicine-oriented framework. A qualitative comparative review was conducted across benchmark datasets, task-specific convolutional and transformer models, domain-adapted vision-language encoders, and recent generative large multimodal models. The evidence shows a clear progression from closed-set classification toward open-ended generation, but clinical deployment remains restricted by small and imbalanced datasets, weak external validation, hallucination risk, limited uncertainty reporting, and insufficient visual grounding. To address these limitations, the study proposes TRACE-MedVQA, a Telemedicine-Ready, Retrieval-Augmented, Calibrated, and Explainable framework. It combines modality-aware image encoding, clinical question encoding, cross-modal attention, trusted knowledge retrieval, dual answer routing, confidence calibration, abstention, visual explanation, and clinician oversight. The framework offers a practical research direction for safer remote decision support while preserving human responsibility.

Cybersecurity education must explain how awareness is associated with willingness to enact protective practices while maintaining a clear distinction between intentions and verified behavior. This study examines a Cisco-oriented awareness–appraisal–intention framework using 385 completed primary-survey records reported as collected from technology users in Bangladesh. Protection motivation theory motivates two parallel appraisal pathways between cybersecurity awareness and secure-practice intention. The analysis combines measurement diagnostics, standardized composite-score regression with heteroskedasticity-consistent inference, 10,000 bootstrap resamples, and ten repetitions of ten-fold cross-validation. Awareness is positively associated with threat appraisal and coping appraisal. In the intention equation, awareness, threat appraisal, and coping appraisal have coefficients of 0.213, 0.197, and 0.273, respectively, with R-squared of 0.294. The two indirect associations are 0.077 and 0.124, but their difference is uncertain. The full model reduces held-out root mean squared error from 0.473 to 0.400 on the original five-point scale. Internal consistency is low across all four scales, and substantial discriminant-validity problems limit confirmatory interpretation. Accordingly, the reported associations are exploratory evidence from the analyzed survey records, not proof of causal mechanisms or actual secure behavior. The findings motivate improved instrument validation, task-specific security measurement, and longitudinal evaluation.