ABSTRACT. Large Language Models (LLMs) are increasingly used to access and interpret information, raising concerns about their social bias and influence on public discourse. Despite growing interest in this issue, political bias in LLMs remains underexplored, particularly in low-resource languages. In this work, we investigate political bias in Polish LLMs and examine whether ideological tendencies can be steered using parliamentary speeches. We construct the Polish Parliamentary Speeches Corpus (PPSC) and fine-tune Bielik-11B-v2.2 and PLLuM-12B-instruct to produce left- and right-leaning variants. To evaluate ideological behavior, we introduce a Political Views benchmark with 266 policy statements across five domains. Our experiments show that both base models exhibit left-leaning tendencies under a basic prompt, that fine-tuning shifts ideological behavior asymmetrically and in a model-dependent way, and that system prompts can influence ideological responses more strongly than fine-tuning.

ABSTRACT. Clinical sentence classification is commonly benchmarked by collapsing expert disagreement into a single majority label and evaluating on sentence-level splits that can leak local context across folds. We revisit this setup for Polish clinical natural language processing (NLP) through PL–ClinDis, a pilot corpus of 100 de-identified discharge-style sentences annotated independently by 20 physicians (2,000 judgments). We evaluate lexical, neural, and lexical–neural hybrid models using grouped three-fold cross-validation across 46 sentence-context groups, thereby preventing local leakage. Term frequency–inverse document frequency (TF–IDF) with logistic regression remains strong on discrimination (0.363 pooled macro-F1), whereas neural models improve probabilistic behaviour: the item-level soft-label domain-adaptive pretraining (DAPT) model achieves the best soft negative log-likelihood (1.232) and Brier score (0.178) in the study and improves expected calibration error over TF–IDF (0.274 to 0.195), although the lowest expected calibration error is obtained by a context-aware neural variant (0.177). A nested hybrid achieves the best macro-F1 (0.439) and shows a paired-bootstrap improvement over TF–IDF, but it worsens calibration. Per-class and significance analyses show that the score ceiling is driven by three minority classes and by genuine, structured annotator disagreement. The study provides a compact evaluation template for disagreement-aware, leakage-safe Polish clinical NLP.

ABSTRACT. Large language models (LLMs) are typically deployed as standalone systems, although many professional tasks require structured collaboration among specialists. This paper presents a proof-of-concept design and preliminary evaluation of a multi-agent LLM framework for multidisciplinary case deliberation, using a simulated medical board as a high-stakes test scenario. A Python-based orchestrator manages specialist roles and agent communication, aggregating their findings into a structured clinical assessment. We apply the framework to four literature-based clinical cases and compare its outputs against reference clinical pathways using embedding-based semantic similarity across six categories: confirmed diagnoses, suspected diagnoses, treatment plans, risks, next steps, and notes. We further compare the multi-agent setup with a single-model baseline. The preliminary results indicate that the multi-agent approach achieves stronger alignment in several clinically relevant categories. Especially when multi-perspective reasoning and risk synthesis is required. Although the advantage is not uniform and the study is subject to important limitations: a very small case set, absence of clinician evaluation, potential training-data contamination from published sources, and the use of an auxiliary LLM for reference extraction. Limited generalization requires further expert-led validation, but the study proves that orchestrating specialized LLMs for expert-style deliberation is feasible.

ABSTRACT. This paper analyzes the transmission of macroeconomic shocks (inflation, interest rates, GDP) to technical provisions (TP) and liquidity (cash demand) of insurers in Poland, Czechia, and Hungary during 2016–2024. Comparing Ridge, Random Forest, Gradient Boosting, and Panel Fixed Effects models, we show that tree-based ensembles outperform linear baselines, confirming the non-linear nature of reserve dynamics. We find a significant strengthening of the “Inflation-Liquidity Nexus” during the 2022–2023 inflation peak as insurers expanded cash buffers. K-Means clustering reveals that business models, not geography, drive solvency behavior. Forward-looking forecasts (2025–2027) under monetary easing project reserve stabilization. These findings highlight the value of machine learning for macroprudential risk monitoring in CEE markets.

ABSTRACT. Modern Text-to-Speech (TTS) systems demand massive training corpora, creating a severe bottleneck for low-resource languages like Polish. Furthermore, relying on commercial cloud-based TTS platforms introduces critical risks to data privacy and latency. This paper presents an effective cross-lingual fine-tuning strategy that adapts a pre-trained, English-centric non-autoregressive architecture (F5-TTS) for high-fidelity, locally deployable Polish speech synthesis. To overcome the open-source data deficit, we utilize a hybrid dataset combining crowdsourced amateur recordings with a proprietary professional studio corpus, which notably includes the deep and distinct voice of the prominent Polish actor Piotr Fronczewski. Objective metrics (UTMOS, WER) and subjective MUSHRA evaluations demonstrate that our fine-tuned model significantly outperforms the current local state-of-the-art baseline (XTTS-v2), achieving highly natural, artifact-free zero-shot voice cloning. Coupled with a lightweight on-premise GUI, this work provides a robust solution for Polish speech synthesis, serving as a blueprint for adapting a foundational TTS architecture to other underrepresented languages.

ABSTRACT. Multi-criteria decision analysis often involves large numbers of criteria, which increases cognitive burden and data acquisition costs. This paper proposes a framework that integrates TOPSIS with LASSO regression to identify a compact, interpretable subset of criteria that preserves the ranking produced by the full model. Unlike PCA-based approaches, the method operates in the original criterion space and retains direct interpretability. The framework is demonstrated on seven Polish offshore wind farms evaluated against 33 criteria. LASSO selects 8 criteria at the cross-validated penalty level, yielding Spearman $\rho = 0.89$ and Kendall $\tau = 0.71$ relative to the reference ranking, with no alternative shifting by more than one position. Bootstrap stability selection identifies a robust core of four criteria, and weight perturbation analysis confirms robustness. The trade-off analysis also shows that smaller retrospective subsets can reach higher rank agreement, which exposes the gap between predictive optimization and rank preservation.

ABSTRACT. We analyze the strategies of companies offering green ICT solutions and thereby driving the dual digital and environmental transition. Building on the “Blue Ocean” strategy, we introduce the concept of a “green network strategy,” which posits that, in the context of sustainable development, it is appropriate to implement open, ecological, and social innovations that will offer new value to customers while optimizing costs. We conducted two surveys in 2024 and 2025 among companies in the IT and service sectors in Europe. We analyzed the results using proportion difference tests and logistic regression. The results show that the surveyed companies developing green ICT solutions are more innovative and more actively engaged in collaborative networks. They exhibit characteristics typical of Blue Ocean Strategy, such as uniqueness and a departure from the industry strategies. They also optimize costs. Engagement in green innovation is associated with better growth prospects and does not reduce competitiveness.

ABSTRACT. This paper comes from our recognition of how Information Systems (IS) are being transformed in the age of artificial intelligence (AI). Information Systems Development (ISD) is no longer exclusively a domain of professional developers. Today’s IS resembles an assembly of tools, technologies, and platforms. Our paper responds to the call for the IS discipline to focus on the ‘IS artifact' by considering IS as an ecosystem consisting of social, information, and technology domains. We couple this with practice theory to articulate the human activity system (HAS) that is enmeshed in IS. We illustrate this approach by ‘zooming in' on data discovery practices of clinical researchers. This re-analysis demonstrates how a focus on existing and emergent practices as ‘practice objects' can better inform ISD. Our contribution is a matrix that combines these two positions to guide ISD that ‘privileges the social' to emphasise a human-centric approach in the context of AI disruptions.

ABSTRACT. Older adults are increasingly targeted by online fraudsters, losing millions of dollars each year to scams in the United States alone. This paper systematically reviews efforts to prevent these scams across multiple domains of research. We apply and expand upon an existing categorization of online fraud prevention efforts to describe the available research. Additionally, we identify and group both the vulnerabilities of older adults targeted by fraudsters and the attacks defended against. Finally, we identify areas for future research in this growing field.

ABSTRACT. This study examines the influence of the external environment on the adoption of Software-as-a-Service (SaaS). Drawing on Theory of Isomorphic Changes and Technology Acceptance Model (TAM), we propose a multifaceted model to evaluate how the external environment, top management support, and IT competencies drive SaaS adoption decisions. The model was empirically validated through a survey among 271 Polish IT practitioners, with data analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). The results indicate that external pressures play a relatively modest role in the SaaS-enabled digital transformation process. In particular, vendor pressure had a negligible direct and indirect impact on the transition to cloud-based solutions. These findings suggest that internal perceptions of technological utility are primary drivers, necessitating a deeper exploration of intra-organizational beliefs. Furthermore, our research highlights the pivotal role of vendor trust in mitigating perceived risks, fostering employee SaaS competencies, and securing executive-level commitment to cloud integration.

ABSTRACT. This article explores how representatives of IT Small and Medium Enterprises (SMEs) understand the Responsible Digitalization (RD). The study is based on 18 in-depth semi-structured interviews and applies an inductive qualitative content analysis to maintain the exploratory character of the research while ensuring analytical rigor and transparency. The results indicate that, although RD is not widely recognized or explicitly used in business discourse, its basic principles are reflected in everyday organizational practices. The analysis revealed a wide range of interpretations of RD, which were subsequently organized into five aggregate dimensions: (1) operational streamlining and resource reduction, (2) human-centered change, (3) inclusion and social value creation, (4) secure and controlled digital environments, and (5) pragmatic business logic. The results suggest that responsible digitalization in SMEs is primarily practice-oriented, performance-driven, and shaped by the specific context in which each enterprise operates.

ABSTRACT. The use of still-developing artificial intelligence (AI) technology is expanding. AI tools are being adopted across various industries, including software development. However, the influence of agility on AI adoption in software engineering remains largely unexplored. The study examines how agility at both the organizational and individual levels affects AI adoption. It uses Structural Equation Modeling (SEM) and surveys of 943 agile software developers from Poland and Finland. The findings reveal that successful AI adoption heavily depends on employees’ agile mindset, which must be supported by an agile organization (AO). An agile organization empowers both employees with agile and non-agile mindsets to promote AI adoption. However, employees with an agile mindset tend to adopt AI more readily than non-agile employees. From a practical perspective, it shows that the IT organization should focus on agility at both the organizational and individual levels to leverage the new capabilities of AI, rather than on rigid processes and policies to limit the potential negative impact of still-developing AI technology.

ABSTRACT. This study investigates whether psychological safety may be at risk during the implementation of the Scaled Agile Framework (SAFe) in a large Swedish insurance company. Using two organization-wide survey waves, one administered at the implementation stage of SAFe and the other approximately 2.5 years later, psychological safety was measured, and mean levels were compared across waves. Results show a statistically significant decrease in psychological safety over time (mean difference = 0.31, p = .02; small-to-moderate effect, d = 0.414). The findings are consistent with concerns that multiteam level coordination and public commitment-making may be associated with higher perceived interpersonal risk. These findings suggest that psychological safety may be part of the “hidden cost” landscape of scaling: an outcome that matters because large-scale software development relies on speaking up about risks and dependencies.

ABSTRACT. Quality Requirements (QRs, also called non-functional requirements) pose a challenge for agile teams, which tend to prioritize functional requirements in their day-to-day practices. In this paper, we investigate web accessibility as an under-researched example of QRs. We conducted a single case study of the implementation of web accessibility in a large public organization that uses agile methods. Our findings, based on interviews and on-site observations, show how web accessibility is addressed at the organizational level and within agile teams. Despite employing several methods and practices, teams still face challenges, including a broad set of web accessibility standards to comply with, a lack of a unified definition of web accessibility within the organization, limited knowledge of the subject among team members, and shortcomings in web accessibility testing tools. These challenges lead the teams to employ ad hoc approaches in their development. The findings are not specific to web accessibility, and we discuss their implications for QRs in general, particularly in large agile organizations.

ABSTRACT. Context: The Spotify Model, introduced in 2012, gained significant popularity as an alternative to formal scaled agile frameworks. Its primary appeal was a unique focus on team autonomy and continuous tailoring, rather than on prescriptive processes. Problem: Despite its widespread influence, a cargo cult 1 adoption has emerged. Objective: This study investigates the evolution of the Spotify Model beyond its 2012 origin, to answer the following questions, (1) how has the Spotify Model been tailored for different industrial contexts? (2) what are the primary challenges, limitations and frictions reported in empirical studies? and (3) what learnings and solutions have the literature identified to overcome these challenges? Method: We conducted a Rapid Review of studies published between 2012 and 2025. The findings presented serve as interim guidance regarding the industry-led evolution of the Spotify Model. Conclusions: The primary learnings from the industry are that (1) tailoring is essential, and it cannot be copied, and (2) organizations must add mechanisms for alignment and governance to balance its focus on autonomy

ABSTRACT. The main goal of this article is to identify the effects of the transformation of the project manager's role in IT project implementation. The diversity of views expressed in the literature on this topic prompts us to present our own position based on our own research, based on surveys, interviews, and our own experience with previously implemented projects. The study results demonstrate the changes that have occurred in recent years during the transition from traditional project management to agile project implementation. Based on the specification of differences, recommendations are proposed regarding the key role of the project manager in the success of agile project management methods in the overall value creation process within a company (product-centric). The study results allow for recommendations regarding the transformation from traditional methods to agile or mixed methods. The study's contribution lies in its mixed-methods research design. It combines a quantitative longitudinal survey tracking shifting expert perceptions between 2017 and 2023 with qualitative in-depth interviews (N = 11) of active IT project professionals to formulate actionable transformational recommendations. The study's findings can be used by both management theorists and practitioners – project managers and users of IT systems.

ABSTRACT. The increasing use of artificial intelligence (AI) in the workplace raises questions about changes in employees’ competence profiles and patterns of AI use. This study examines these changes using a comparative approach based on two waves of survey data collected in Poland at the turn of 2024/2025 (N = 288) and 2025/2026 (N = 308). The same research instrument was applied in both waves, providing direct comparison over time. The results show an increase in the use of AI tools in professional work, while limited differences in declared competence levels were observed between the two waves. Additional analyses reveal that the relationship between competencies and AI use varies across different areas, and that distinct user profiles can be identified based on the combination of competence levels and AI usage patterns. The findings suggest that changes related to AI adoption follow a gradual and adaptive pattern rather than a disruptive one. The study contributes by providing a short-term comparative perspective on competence development in AI-supported work environments.

ABSTRACT. This study proposes a novel deep learning method for remote sensing imagery domain adaptation for the purpose of building vectorization. The method, which uses a Generative Adversarial Network to extract chromatic data from the target image and apply it to the source imagery tiles, was separately trained on the SpaceNet AOI 2 - Las Vegas dataset as well as the SpaceNet AOI 10 - Dar Es Salaam image and applied to process the CrowdAI dataset, which was then used to train the state-of-the-art HiSup building vectorization network. Finally, the SpaceNet AOI 2 - Las Vegas and a dedicated Dar Es Salaam evaluation dataset were used to test HiSup’s vectorization performance. When tested on Dar Es Salaam, the proposed method improves upon state-of-the-art domain adaptation approaches such as ColorMapGAN, Reinhard color transfer and CycleGAN by 17% in Average Precision, 14% in Average Recall, 7.2% in Intersection over Union (IoU) and 6.4% in Complexity Aware IoU (C-IoU). A combination of existing methods was outperformed by 14% in AP, 9% in AR, 4% in IoU and 4% in C-IoU.

ABSTRACT. The article proposes a new method for user recognition based on their unique eyelid blinking pattern. In our study, we utilized the publicly available mEBAL database and the data we collected. The temporal eyelid movement patterns extracted from the samples in the database are analyzed by a siamese neural network. Our research aimed to develop a method that is resistant to shoulder surfing and brute force attacks, while also not requiring complex recording devices. Most user authentication methods utilizing eyelid blinking patterns are vulnerable to pattern replication attacks. The achieved results of 98.20% accuracy and 0.11 EER unequivocally demonstrate the superiority of the proposed method over other methods using eyelid blinking for user authentication.

ABSTRACT. This project investigates the use of self-hosted Large Language Models (LLMs) for processing Polish medical text within a GDPR-compliant, on-prem environment. The goal is to assess whether relatively small, locally deployed LLMs can be used in tasks such as data extraction and translation of Polish medical records. We tested our solution on anonymized pediatric data set of 10 years of activity in Tuchola Hospital (Szpital Tucholski). We evaluated the performance of small and medium quantized models on translation and extraction tasks, measured the energy efficiency, and validated the quality of model generated output against ground truth. Experiments revealed that translation and extraction quality depend on the quality of the source text, and using the biggest model possible is not cost-optimal in terms of used energy vs. end accuracy, with smaller models with extra context performing up to 7pp. better than bigger models for translation and 25pp. for extraction.

ABSTRACT. This study investigates whether investment in information and communication technologies (ICT) results in measurable household energy savings and reduced energy poverty across the European Union. Drawing on panel data for 27 EU countries over 1995–2024, a first-difference model with time fixed effects and Driscoll–Kraay robust standard errors was employed. At the EU-wide level, no statistically significant relationship was found between ICT expenditure and the household energy saving rate, energy poverty rate, or final energy consumption. National-level estimates, however, indicate different findings: investments in computer hardware and telecommunications significantly affect the heterogeneity of ICT effects, although to a limited extent. Country-level heterogeneity mainly concerns the impact on energy poverty. The absence of broader effects — especially in lower-income countries — points to rebound mechanisms and the rising energy intensity of ICT technologies themselves.

ABSTRACT. Smishing--phishing conducted via SMS--continues to spread globally. Most detection systems are built using only English data, which limits their use for other languages. We propose a multilingual smishing detection framework based on self-training with XLM-RoBERTa. Starting with English-labeled data, we apply zero-shot inference to Bengali and Swahili messages, extract high-confidence predictions, and incorporate them as pseudo-labeled examples for fine-tuning. Our approach improves recall and F1 scores in both target languages without requiring parallel corpora or costly annotations. We further analyze the impact of language-specific and balanced pseudo-label augmentation through ablation studies. The results show that the combination of samples from multiple languages leads to better generalization and reliability. This work highlights an effective low-resource strategy for building multilingual smishing detectors, allowing a broader deployment across linguistically diverse user populations.

ABSTRACT. Integrity and non-repudiation of operational messages exchanged between logistics operators are critical for dispute resolution and regulatory compliance. This paper presents TDI (Trusted Document Infrastructure), a blockchain-based system that adds a tamper-evident authentication and audit layer on top of existing message exchange infrastructure, without modifying legacy systems. TDI combines bidirectional ECDSA signatures (the sender signs to prove authorship, the receiver signs to establish non-repudiation of delivery) with smart contract storage on an EVM-compatible blockchain. A configurable dual-storage strategy assigns on-chain or off-chain storage by message criticality, balancing security against cost. We implemented and evaluated TDI in a Portuguese port and railway logistics project using real data (approximately 41,000 messages per month). The mixed strategy reduces on-chain storage cost by 77%. Validation on both public Polygon and private Hyperledger Besu networks confirms portability across EVM platforms. TDI lets operators ensure message authenticity and accountability without disrupting established workflows.

ABSTRACT. The current status quo of explainable AI makes use of prefabricated explanations to enable users to gain insights into an AI-based system’s underlying decision-making logic, thereby increasing users’ system understanding. In contrast, we introduce an established method from the field of generative learning called self-explanation: Rather than passively processing existing explanations, the recipient is integrated into the explanation generation process and (to different degrees) generates their own explanations. We performed an experiment to investigate the effects of such self-explanations (with various levels of contextual information) on system understanding (operationalized through explainability, transparency, predictability) and cognitive load. Our findings suggest that recipient-generated self-explanations can be effective at increasing system understanding, providing an effective alternative to the current status quo with comparable mental load.

ABSTRACT. Graph kernels are a powerful tool for analyzing graph datasets. Most existing approaches focus solely on optimizing performance metrics rather than model explanation quality. In this paper, we introduce the Generalized Graphlet kernel, an extension of the classical graphlet kernel. Unlike its predecessor, it optimizes both classification accuracy and the quality of model explanations. The resulting Pareto-optimal subset of graphlets becomes an explanation for the model, making the model itself a source of knowledge, not just the input dataset. We evaluate our method on 12 benchmark datasets for biochemical graph classification using AdaBoost, Random Forest, and Support Vector Machine classifiers. Our results show that our method consistently improves explanation quality and classification accuracy compared to the standard graphlet kernel, particularly for SVC and AdaBoost. The Pareto-optimal graphlets provide interpretable, domain-relevant patterns, demonstrating that our method creates models that are both accurate and explainable.

ABSTRACT. This article proposes an Adversarial Prototype Decomposition (APD) algorithm as a method for improving the scalability of the Support Vector Machine (SVM) classifier. The algorithm partitions large datasets into smaller, more manageable subsets called regions, enabling the application of SVM classification to datasets containing hundreds of thousands of samples. A series of experiments were conducted on medium- and large-scale datasets to evaluate the approach. For medium-sized datasets (up to 60,000 samples), the algorithm maintains the same predictive performance as the reference SVM while accelerating the training and prediction phases by an average factor of 2.5. On large-scale datasets containing 500,000 samples, the proposed APD-SVM solution achieves significantly higher performance than the reference Cascade SVM, while simultaneously speeding up the training process by 1.7 times and the prediction phase by up to 36 times. Furthermore, the experiments demonstrate the importance of the Learning Vector Quantization (LVQ) procedure in optimizing the prototypes within the APD framework.

ABSTRACT. This paper develops an interpretable decision-support framework for speedway analytics based on an existing Expected Points (xP) model. The proposed framework adds a two-level interpretive layer to model outputs. First, heat-level entropy is used to describe the dispersion of normalized expected-point shares within a heat. Second, rider-level Bias and the Rider Deviation Index (RDI) quantify the direction and variability of longitudinal deviations between observed and expected performance. The empirical analysis is based on six seasons of Polish PGE Ekstraliga data (2020-2025), comprising 23,500 rider performances across 406 matches. The results indicate that entropy distinguishes heats with stronger structural imbalance from those with more even competition. In contrast, Bias and RDI identify riders with systematic overperformance, underperformance, or stable deviation patterns relative to model expectations. The study contributes a formal interpretive framework for translating existing xP outputs into decision-oriented summaries at both the heat and rider levels.

ABSTRACT. Feature selection remains essential in machine learning, but traditional wrappers like Recursive Feature Elimination (RFE) suffer from quadratic complexity (O(d²)) and susceptibility to data leakage. We propose the Adaptive Interval Feature Selector (AIFS), a non-iterative wrapper-style selector grounded in aggregation functions. By mapping tree-based feature relevance scores into stability intervals across localized folds, AIFS constructs a global optimization sequence without sequential retraining, scaling near-linearly at O(M · d · N log N). Crucially, AIFS isolates feature selection from validation partitions to prevent data leakage. Empirical evaluation across 25 datasets demonstrates that AIFS achieves robust performance, yielding an average Balanced Accuracy of 0.8334 (outperforming standard wrappers), highly stable subset consistency (average Kuncheva Index of 0.6904), and resilience to 30 % missing data. Theoretical complexity scaling projects speedups of up to five orders of magnitude in ultra-high-dimensional spaces, successfully bridging the gap between filter efficiency and wrapper discriminative power.

ABSTRACT. The study aimed to identify employee profiles reflecting combinations of quiet quitting, passive quitting, and work engagement. Using a person-centred approach and unsupervised learning, survey data from 1,040 employees were analysed. Clustering relied on composite indices derived from abbreviated quiet and passive quitting scales and the Utrecht Work Engagement Scale-9 (UWES-9). Multiple algorithms (k-means, hierarchical clustering, spectral clustering, Gaussian mixture models) were compared, and the optimal solution was selected using separation metrics (Silhouette coefficient, Davies–Bouldin index, Calinski–Harabasz index), information criteria (Bayesian Information Criterion [BIC], Akaike Information Criterion [AIC]), and bootstrap stability (Adjusted Rand Index [ARI]). A four-profile solution was selected as the most parsimonious and interpretable configuration across the compared models. The profiles differed mainly in boundary regulation, exhaustion-related withdrawal, and work-related energy, although the separation and stability indices indicate that the solution should be interpreted as moderately stable rather than definitive. Findings suggest quiet quitting and passive quitting are related but distinct withdrawal mechanisms. The study advances profile-based research on employee withdrawal and highlights implications for targeted human resources (HR) interventions.

ABSTRACT. This study examines the influence of AI-enabled Intelligent Transportation Systems applications and logistics ambidexterity on green logistics practices and their ramifications for both green innovation and digital transformation. This study also analyses the impact of green innovation on the process of digital transformation in freight transport organisations. Empirical evidence was collected from 246 freight transport enterprises operating in Poland through the application of a structured survey methodology. The acquired data were assessed using descriptive statistics and the Structural Equation Modelling methodology. The study offers empirical evidence for an integrative perspective that foregrounds intelligent technologies as drivers of the emergence of green logistics practices and their implications for innovation and digital transformation.

ABSTRACT. Public procurement plays a pivotal role in public service delivery and strategic governance. Growing demands for its transparency, efficiency, and responsiveness have encouraged the exploration of AI in the public procurement process. AI offers potential benefits, including improved efficiency, enhanced effectiveness, greater automation in transactions, and streamlined supplier management. Despite these benefits, its adoption in public procurement is in its early stages. Existing research has largely focused on the discussion around the potential of AI in procurement processes, whereas limited attention has been paid to the determinants of AI adoption in public procurement. This paper conducts a Systematic Literature Review (SLR) to identify the determinants influencing AI adoption in public procurement. The findings are analysed using the Technology-Organisation-Environment (TOE) framework. The results reveal that organisational determinants vary across the procurement process, while the TOE dimensions differ across stages of the adoption process. Public value also emerges as an important dimension.

ABSTRACT. This study examines the complex interplay between AI familiarity, trust in AI, innovation orientation, change acceptance, and fears of job replacement among employees in the financial sector. Drawing on a sample of 196 respondents and employing Partial Least Squares Structural Equation Modeling, the analysis reveals that AI familiarity significantly predicts trust in AI (β = 0.463, p < 0.001), which in turn positively influences innovation orientation (β = 0.609, p < 0.001). However, neither change acceptance nor innovation orientation demonstrated significant effects on replacement fears, and trust in AI did not significantly alleviate such concerns. These findings suggest that employees cognitively distinguish between technological appreciation and employment security, indicating that trust in AI as a functional tool does not translate into reduced anxiety about workforce displacement. The study contributes to the literature on technology acceptance, organizational change, and the future of work by highlighting the complex, non-linear relationships between cognitive, affective, and behavioral responses to AI implementation. Theoretical implications for understanding AI as an ambivalent technology and practical recommendations for organizations seeking to foster innovation while managing employee concerns are discussed.

ABSTRACT. E‑services play an increasingly critical role in Higher Education Institutions (HEIs), supporting core academic, administrative, and governance functions. Despite their strategic importance, limited research has systematically examined how e‑services are assessed and improved from an institutional perspective. This paper presents a Systematic Literature Review (SLR) to identify instruments proposed or applied to evaluate and enhance e‑services in HEIs. A search across four major academic databases yielded 573 articles, of which only six were relevant. The findings reveal a fragmented body of literature, largely dominated by user‑centric and quality‑based approaches, with very limited adoption of maturity or capability perspectives. The review highlights a significant gap between the relevance of e‑services for HEIs and the availability of structured assessment and improvement instruments. Based on these findings, the paper outlines directions for future research, emphasizing the need for a multidimensional and modular e‑service maturity model tailored to public HEIs.

ABSTRACT. This study investigates enterprise information systems development (ISD) in the European Union from a multidimensional perspective, examining the roles of industrial sector and firm size in shaping digitalisation patterns. Using harmonised Eurostat data on ICT usage in enterprises and principal component analysis (PCA), the study identifies latent dimensions of ISD and compares their structure across sectors and enterprise size classes. The results reveal that firm size strongly influences overall digital intensity, reflected in a clear and monotonic gradient in digital maturity. Sectoral affiliation contributes more strongly to multidimensional variation in ISD structures, although the relative importance of sector and size depends on the specific digitalisation dimension considered. The findings indicate that enterprise digitalisation in the EU is structurally differentiated and jointly shaped by organisational scale and industry context rather than following a uniform development path.

ABSTRACT. AI coding assistants are used daily by most professional developers, yet organisations treat them as individual tools within stable team structures, leaving fragmentation and governance unaddressed. This study investigates whether practitioners perceive AI integration as an organisational design problem and whether they consider a three-layer model (System, Platform, and Product Teams) plausible. We surveyed $n=99$ professional developers with 28~Likert items normalised to a $-3$ to $+3$ scale. Practitioners endorse AI development environments as shared infrastructure, yet a gap between the model's abstract plausibility ($M=+1.49$) and its perceived applicability ($M=+0.76$) reveals an acknowledged maturity deficit. Intensive AI users frame integration as an organisational challenge more strongly ($d=1.63$, $p<.001$) and report pronounced shifts from producing code towards reviewing and steering AI outputs ($r=.790$, $p<.001$). Agile practices evolve selectively, with Continuous Integration and TDD gaining governance significance. Organisations therefore have a window of opportunity to respond structurally before fragmentation becomes entrenched.

ABSTRACT. This study investigates how different narrative framings of Digital Twin technology influence moral acceptance and risk perception in a tourism context. Drawing on an experimental vignette design, respondents were randomly assigned to one of three scenarios (environmental protection, infrastructure optimization, and data-driven monitoring), followed by scenario-specific informational cues. Results show that exposure to the monitoring-oriented narrative was associated with a greater decline in moral acceptance than exposure to environmental- and infrastructure-oriented narratives, while increasing perceived risks related to privacy and freedom. Privacy concerns emerge as the strongest and most consistent predictor of declining acceptance, whereas perceived freedom violations show weaker but related effects. No significant moderating effects of user status (resident vs. tourist) were observed, suggesting a shared cognitive response pattern across user groups. From a pedagogical perspective, the study highlights the role of communicative framing in shaping the legitimacy of emerging digital infrastructures, such as Digital Twins, in public-sector tourism governance.

ABSTRACT. The development of modern technologies, including artificial intelligence (AI), currently has a significant impact on higher education and students studying at HEIs (Higher Education Institutions). The aim of the study presented in this article was to examine patterns of AI tool usage and students’ perceptions of their role in the educational process among students from Poland and Ukraine. An exploratory study was conducted on a sample of 108 students, including 54 from each of the analyzed countries. To identify relationships between selected variables, Spearman’s rank correlation and Pearson’s chi-square tests were applied based on survey data. The results revealed significant relationships between the analyzed variables and provide a basis for further comparative research on the use of AI in higher education.

ABSTRACT. Public organizations are undergoing profound AI-driven transformation, creating an urgent need for competencies that integrate technical, organizational, and societal perspectives. Existing educational programs primarily target technical specialists, leaving a gap for domain experts responsible for leading and managing AI initiatives. This study adopts a Design Science Research (DSR) approach to develop a master’s program for non-technical public-sector professionals. The program is grounded in a sociotechnical information systems perspective and structured around the AI system development life cycle. Its design is informed by literature on AI in public organizations, policy and industry reports, and empirical data from workshops, discussions, and interviews with key stakeholders. The program blueprint was evaluated through comparison with existing AI-related programs, alignment with the WASP-ED curriculum, workshop discussions, and expert review. The study contributes by identifying and structuring competence needs for AI-driven transformation and translating them into an educational artifact that supports organizational digital transformation.

ABSTRACT. Youth remain an underexplored stakeholder group in Information Systems (IS) research. Educating them for their digital future should be a pivotal concern of ours. This study, inspired by the Participatory Design (PD) tradition, explores ways by which to offer empowerment and impact-oriented IS education. Literature on IS education lacks PD inspired approaches. We conducted ten participatory game design workshops with youth, who designed and evaluated a literacy game for youth. Their contributions shaped multiple aspects of the game: character design and visual elements were implemented directly, while other ideas evolved into new features or inspired them. Their input also influenced game mechanics and narrative flow, and some ideas travelled beyond the initial project, informing virtual reality environments. We demonstrate an example of PD inspired, empowerment and impact-oriented education, which enabled youth to impact not only immediate design outcomes but also future development trajectories.

ABSTRACT. Recent advances in artificial intelligence (AI) present innumerable opportunities for development of learning tools for higher education. This paper explores the potential for AI to augment scenario-based learning (which we have termed "a-SBL") with a particular focus on supporting professional-oriented outcomes for students of human-centered computing (HCC). This paper presents an exploratory period of "reconnaissance work" between Allied Health (AH) and HCC educators conducted to inform an empirical study of AI enabled embodied conversational agents (ECAs) in both programmes. By documenting our interdisciplinary, co-learning journey, we argue that, despite realisable potential via emerging platforms to create and employ such ECAs, effective deployment requires consideration of sociotechnical models that are educator-informed. We propose the concept of a 'maturity assessment' to highlight potential opportunities and limitations at the curriculum level ahead of adopting AI tools with SBL activities. Our explorations aim to inform others working toward a-SBL approaches, but note that further work is required to fully examine long-term benefits and the ethical principles for use of AI enabled tools in higher education.

ABSTRACT. Generative AI has changed the way we learn and teach introductory computer programming today at the university. This transformation presents many challenges, problems and opportunities that require attention from educators: (1) challenges of academic lecturing process, (2) teaching improvement of integration AI tools to introductory programming course. As these transformations are most pronounced at the introductory level of programming education, it is crucial to develop competent students who can responsibly use AI tools in their learning process. Based on our extensive experience in teaching programming in higher education and conducted a survey-based investigation of introductory programming course students, we present insights and suggestions for AI tools integration. Our research, based on student's applied AI interaction strategies performing programming tasks and how it influences experienced cognitive issues interaction according to AI tools.

ABSTRACT. This article presents the results of a study conducted at the Faculty of Management, University of Gdańsk, Poland. The aim of the study is to examine students’ perceptions of the university’s role in developing digital competencies among accounting students. Four research questions were formulated concerning the primary sources of knowledge and the perceived development of digital competencies among the respondents. The study combines a literature review, a survey, and an analysis of curricula and course syllabi. A total of 258 valid responses were collected from undergraduate, graduate, and postgraduate Finance and Accounting students, yielding a response rate of 64.8%. The findings indicate that students primarily develop digital competencies during their studies in basic areas such as spreadsheets and word processing. In contrast, competencies related to accounting software and artificial intelligence remain underdeveloped. The curriculum analysis confirms that artificial intelligence tools are not integrated into the teaching programs. The results suggest a gap between current educational offerings and the digital skill requirements of the accounting profession. The study is limited to a single institution and field of study; therefore, the findings cannot be generalized but provide a basis for further research on digital competence development in higher education.

ABSTRACT. This paper introduces a novel strategic optimization methodology for Uninorm Petri Nets tailored for intelligent information and decision support systems operating under uncertainty. The proposed hybrid framework, termed Optimized Uninorm Petri Nets, bridges the gap between theoretical modeling and practical system development by integrating algebraic optimization with simulation-based verification. The methodology is built on two strategic pillars crucial for modern information systems engineering. The first pillar, strategic logic optimization, utilizes a mathematically constrained selection of uninorm operator triplets (In, Trs, Out) to ensure the structural reliability, precision, and efficiency of the system's core decision-making logic. The second pillar focuses on dynamic system alignment through a simulation-driven analysis of the uninorm neutral element e in (0, 1), which optimizes system dynamics and directly embeds a "transparent-by-design" philosophy aligning with Explainable AI standards. Finally, the methodology's utility for information systems development is validated through a practical implementation in a real-world technical control system. Comparative analyses demonstrate that Optimized Uninorm Petri Nets offer superior adaptability, precision in modeling transitional states, and architectural robustness compared to traditional fuzzy frameworks, proving to be a highly effective tool for engineering next-generation intelligent information systems.

ABSTRACT. Predicting the health of software projects is a critical challenge in Information Systems Development (ISD). This paper presents an automated nine-stage pipeline for project health classification that integrates static code metrics with behavioral repository analytics collected via the GitHub API. A labeled dataset of 777 open-source repositories across eight language groups is constructed from 2,113 candidates on a realistically imbalanced class distribution (37% healthy / 63% at-risk), with rule-based labeling applied with explicit leakage mitigation through ablation study. Four ML classifiers are evaluated under 5-fold cross-validation with SHAP explainability and McNemar statistical testing. XGBoost achieves F1 macro = 0.947 ± 0.016 and AUROC = 0.984, significantly outperforming Logistic Regression (p = 0.0002) and SVM (p < 0.0001). SHAP analysis reveals that behavioral metrics dominate static code metrics as health predictors by a ratio of 106:1. The model generalises across all eight language groups (F1: 0.916–0.979), enabling explainable, AI-driven decision support for IS project portfolio management and proactive governance.

ABSTRACT. This paper considers the following problem in multi-agent data analysis: given a tuple of k-valued decision tables (a distributed decision table), construct an approximate shared test of specified accuracy and minimum cardinality that can be applied simultaneously to each of the tables. Because this problem is NP-hard, new theoretical results related to the accuracy of the proposed algorithm were reported, and experimental verification was conducted for the datasets from the stylometry domain. The experiments were carried out taking into account the perspectives of knowledge representation and performance of classifiers for the proposed algorithm.

ABSTRACT. Multi-criteria decision support systems (DSS) rely on criteria weight elicitation, yet the available methods differ substantially in the input they require and the cognitive effort it imposes. The Analytic Hierarchy Process (AHP) requires a full pairwise comparison matrix, the Best-Worst Method (BWM) reduces this to two reference vectors, while Ranking Comparison (RANCOM) requires only an ordinal ranking. These formats differ in how much of a decision-maker's preference they can represent, raising a design question: how much weight-recovery accuracy is sacrificed by adopting a lighter input format? We address this with a large-scale Monte Carlo study in which reference weight vectors are sampled from a Dirichlet distribution, encoded into each method's input, and recovered with its solver. Because the simulated decision-maker is perfectly consistent, the experiment isolates the representational loss of each method rather than the error real experts introduce. We evaluate recovery with four complementary cardinal and ordinal metrics. AHP yields the lowest cardinal error and BWM the second lowest across the whole criteria range, while RANCOM reproduces the input ranking exactly, since it receives that ranking as input. The accuracy gap between methods narrows as the number of criteria grows, while the judgement count and input complexity required increase.

ABSTRACT. The issues raised in the paper concern knowledge discovery and representation by decision rules induced while changing the nature of input domain by transformations. These topics are relevant to computational science because they address fundamental tasks of data processing and analysis and contribute to the search for advantageous data representation. With real-world problems described by attributes of varied type, their transformations such as discretisation can have a significant impact on the effectiveness of decision-making. When data exploration involves induction of decision rules, even when an algorithm can operate directly on the continuous domain, such as MODLEM which was used in the research, discretisation can be used to simplify data and to find such transformation conditions that lead to improved performance.

ABSTRACT. Software Fault Prediction (SFP) is an important task of software engineering, which attempts to discover the faulty modules in source code proactively to improve quality and reducing maintenance costs. The machine learning (ML) models has significantly enhanced fault prediction capability, yet these methods are black box, and limits in interpretability and trust. Explainable AI (XAI) approaches, such SHapley Additive exPlanations (SHAP), are useful to get insights, and quantify the impact of specific characteristics to the model decisions. The study proposed a robust and interpretable framework for SFP which integrates class imbalance handling (SMOTE), optimisation (using hyperparameter tuning (SMOHY)), SHAP-based explainability and Feature Sensitivity (FS) analysis to produce accurate and transparent models using 41 ML classifiers against 41 opensource java projects. SMOHY improved the efficiency and predictability of the SFP model by 22.38% compared to ORGD and 6.5% compared to SMOTE, attaining highest median AUC value of 0.82. ExtraTrees and RandomForest algorithms achieve the highest median accuracies of 87% and 86%. The SHAP values revealed RFC as the most significant feature, offering a visual depiction of its influence on model predictions. FS analysis demonstrated that RFC metrics are the most important metric affecting SFP model performance, validating SHAP’s applicability in SFP.

ABSTRACT. This paper develops a data-driven framework for analyzing irregular circular economy cargo flows in seaports. A composite instability measure, the Intra-Port Irregular Circular Cargo Index (ICCI), is introduced to capture variability and irregular occurrence of cargo streams over time. The index is combined with cargo share to construct a decision-oriented classification matrix linking cargo behavior to management implications. The empirical analysis is based on cargo handling data for six Polish seaports over the period 2011-2022, covering 11,032 data cells. The results show that circular cargo flows are predominantly irregular and low in share, yet operationally demanding. The study proposes a methodology designed for integration into information systems, enabling continuous, real-time decision-making based on observed cargo flow dynamics. The proposed framework enables differentiation between stable, adaptive, and volatile cargo segments and supports more effective, data-driven decision-making in port management under conditions of uncertainty.

ABSTRACT. Tourism’s global economic scale demands coordinated sustainable management that destinations struggle to deliver. The bodies tasked with tourism orchestration, Destination Management Organizations (DMOs), lack the Information Systems (IS) needed to coordinate sustainable practices. At the same time, destination stakeholders differ radically in digital capability, creating what is termed the multi-stakeholder digital literacy gap. This paper aims to design a tourism IS architecture in which autonomous AI agents absorb stakeholder heterogeneity, enabling sustainable coordination without requiring uniform digital literacy or sustainable behavior. Following a Design Science Research (DSR) approach, the study identifies operational challenges from literature, derives design requirements, and translates them into a three-layered architecture based on Multi-Agent Systems (MAS) coordinated via the Contract Net Protocol (CNP), featuring a Triple Sustainability Gate to enforce environmental, social, and economic constraints. The architecture is demonstrated ex-ante against the Taormina-Peloritani area (Eastern Sicily, Italy), drawing on first-hand institutional knowledge from a nascent DMO.

ABSTRACT. The aim of this article is to identify potential areas for AI use in the maritime economy and examine if they are implemented in business. After literature review, the level of implementation of AI solutions was assessed based on a survey distributed between industry representatives. According to the results, the use of the term of AI is often overused, as it tends to be limited to digitalization as well as AI solutions currently support mostly common tasks, and their application to complex decision-making problems is still in the testing phase.

ABSTRACT. This paper examines the determinants of perceived effectiveness of AI tools in sales and marketing among micro businesses and explores how AI adoption patterns and perceptions evolve over time. A questionnaire-based survey of solopreneurs and micro-business owners from Central and Eastern Europe, Western developed markets, and emerging markets was conducted in two waves (2025 and 2026). Descriptive statistics, non-parametric tests, and ordinal logistic regression were used for data analysis. The study investigates the role of behavioral factors, including usage frequency and breadth of application, alongside control variables such as revenue, as well as temporal and regional effects. Results show that the perceived effectiveness of AI tools in sales and marketing is primarily driven by usage intensity, with significant regional differences observed. Furthermore, comparative analysis between 2025 and 2026 reveals a stable perception of the effectiveness of AI tools over time. The findings contribute valuable insights into AI adoption in the micro-business context by demonstrating that perceived value is primarily driven by usage behavior rather than by differences in firm resources, breadth of application, or temporal factors. The study offers implications for entrepreneurs, educators, and policymakers seeking to foster AI adoption and capability development in micro-business contexts.

ABSTRACT. This study investigates the role of neuroscience-inspired tools in decision support within the context of Logistics 5.0. Unlike prior research focusing on individual technologies, this study identifies interdependencies and interaction effects between these tools in user perception. A quantitative survey (CAWI) was conducted on a sample of 423 respondents in Poland. The analysis focuses on selected neuroscience tools, including neuroleadership, neuromarketing, and neuroadaptive decision support systems. The results reveal that the perceived usefulness of these tools is not independent but shaped by complex, non-linear relationships, where certain tools enhance the evaluation of others. Additionally, familiarity with specific tools significantly influences their perceived value. The findings contribute to understanding human-centered digital transformation, highlighting the need to treat neuroscience tools as an integrated ecosystem supporting managerial decision-making in Logistics 5.0.

ABSTRACT. The aim of the article is to identify the factors explaining the use of online insurance among users of digital financial services, with particular attention to the relationship between this service and the extent of FinTech use, as well as perceived digital capability. The study is based on data from a CAWI survey (N = 840) conducted in Poland in 2025 and applies logistic regression models. The results indicate that online insurance is used by a substantial share of respondents, although less frequently than basic FinTech services such as mobile payments or banking applications. Significant differences are observed by age and education level. Online insurance use is significantly associated with both the extent of FinTech use and perceived digital capability, with a more consistent association observed for the extent of FinTech use. The findings indicate that, in the context of digital transformation, online insurance systems should account for differentiated user profiles and accumulated digital financial experience.

ABSTRACT. Autonomous vehicles (AVs) increasingly rely on advanced sensing, Machine Learning (ML), and AI to operate in complex, dynamic environments. Rapid technological change, environmental variability, and evolving regulatory demands create volatility in functional and non-functional requirements, challenging conventional development processes. This exploratory qualitative study examines how requirements evolve in Autonomous Vehicle Perception (AVP) systems and the implications for engineering practice, based on semi-structured interviews with 16 professionals from international automotive companies and research institutes. Findings show functional requirements shifting toward AI-enabled perception, real-time processing, advanced sensor fusion, improved localisation, and region-specific adaptations, while non-functional requirements, including safety, cybersecurity, robustness, reliability, transparency, and scalability, grow more demanding under regulatory, operational, and societal pressures. This volatility creates challenges including sensor uncertainty in adverse conditions, rising development and computational costs, greater compliance burdens, and increased complexity in real-time decision-making. We examine how organisations respond through adaptive practices: Scrum supports rapid ML experimentation, Kanban manages continuous data and annotation workflows, SAFe enables large-scale cross-organisational coordination, and hybrid Agile approaches combined with V-model or Waterfall elements preserve safety, traceability, and compliance. The study contributes empirical insight into adapting Agile and hybrid practices to manage evolving requirements in safety-critical, AI-enabled AV perception development.

ABSTRACT. As agile teams increasingly rely on AI-generated outputs, organizations tend to prioritize explainability improvements to achieve effective governance outcomes. This paper argues that such responses are misdirected and reframes AI governance in agile delivery as an interpretive practice. Enterprise AI governance typically focuses on allocative governance practices, including decision rights, policy constraints, and audit mechanisms, leaving the interpretive governance, constituted through collective sensemaking, unaddressed. The study applies fuzzy-set Qualitative Comparative Analysis (fsQCA) on data from 143 agile practitioners to test three structural tensions that specify distinct governance challenges posed by AI-generated signals. The study identifies the deliberative team core, a set of conditions forming the primary sufficient configuration for distributing governance and sustaining adaptive capacity, with psychological safety as the most calibration-stable constituent; its absence could drive governance failure. The paper presents interpretive governance as a theoretically distinct concept and provides the initial empirical anchors for a multi-regime research program.

ABSTRACT. Public organizations are increasingly adopting agile software development methods (ASDM) despite bureaucratic structures that both enable and constrain agile software development (ASD). This study aims to understand how agile coaches navigate rule-based environments by adapting practices, addressing challenges, and managing tensions between ASDM and public sector governance. A qualitative, deductive approach grounded in New Institutional Theory (NIT) is applied. Data were collected through semi-structured interviews with agile coaches in Nordic public organizations and analyzed using reflexive thematic analysis. Findings show that coaches adopt a situational approach, incrementally adapting practices to organizational structures, culture, and regulatory conditions. Challenges include centralized decision-making, administrative constraints, and risk-averse cultures. Coaches address these through strategies centered on dialogue, language adaptation, and trust-building, enabling gradual transformation within existing structures. The study highlights how institutional conditions shape ASDM and positions agile coaches as key actors in translating and embedding ASD practices in bureaucratic contexts.

ABSTRACT. Conceptual models are often underused in agile contexts. Large Language Models (LLMs) open new avenues to automatically derive such models from agile requirements artifacts. This could potentially reduce documentation effort and reconnect agile practice with model-driven engineering. This study investigates whether LLMs can transform user stories and BDD scenarios into UML use case diagrams and BPMN workflows of sufficient quality for professional requirements engineering practice. Using GPT-o3, the research explores four factors affecting LLM-generated models: prompting strategy, output consistency, comparative model quality, and the impact of expert feedback. Evaluation employs the Lindland–Moody quality framework across syntactic, semantic, pragmatic, and completeness dimensions. As an exploratory feasibility study, results demonstrate promising automation potential while highlighting current limitations in precision, consistency, and pragmatic adequacy.

ABSTRACT. This paper examines the adoption of Scrum framework within a single software development team operating in a non-technology organization, highlighting the challenges and obstacles encountered throughout the process. Using an action research approach, the authors investigate how targeted interventions influenced the Scrum implementation process. The study focuses specifically on team organization, including the structuring and coordination of the development team, as well as the assignment of the Product Owner and Scrum Master roles. As a single-team case study, the research does not aim for statistical generalization; instead, it offers an in-depth, context-rich account from which transferable insights can be drawn. The findings culminate in a set of practical recommendations for software development teams seeking to adopt Scrum. Rather than universally applicable rules, these recommendations are best understood as lessons grounded in the studied context that may inform other enterprises operating under similar conditions and aiming to leverage Scrum to improve productivity and software product delivery. The study further serves as a basis for future multi-team and comparative research that could test the broader applicability of these findings.

ABSTRACT. TAmail is a system and a service that enables sending of anonymous e-mails with the possibility of being responded to. Because TAmail aims at being broadly adopted as a publicly available, operational service, a study that evaluated its applicability potential was carried out. This paper presents the results of the analysis.

ABSTRACT. Context: The adoption of microservice architectures is often driven by organizational benefits like team autonomy and independent deployability, which align well with agile methods. However, this choice is frequently contested due to concerns about performance overhead from inter-service communication. Objective: This paper empirically investigates this performance trade-off across different technology ecosystems. We present a comparative analysis of monolithic and microservice architectures implemented in Java (Spring Boot), .NET (ASP.NET Core), and Python (Flask/FastAPI). Method: A containerized route-planning benchmark with identical REST APIs was implemented in six variants. Each variant was subjected to controlled load tests using Locust, with performance evaluated based on response time percentiles (p50, p95) and throughput. Results: Our findings reveal that for Java and .NET, the choice of architecture has a negligible impact on performance. In contrast, the Python implementation exhibited severe performance degradation as a monolith, though regardless of the architectural choice, remained significantly slower than Java and .NET. Conclusion: These results challenge the axiom that microservices inherently introduce a significant performance penalty. This finding suggests that agile teams can adopt microservices to leverage key organizational advantages without being hindered by performance concerns.

ABSTRACT. This poster introduces Cx, an expert-based composite index for assessing ICT-related risk in adversarial organizational contexts. The index estimates an organization's defensive capability against cyber adversaries by combining technical, organizational, and attacker-oriented factors in a weighted scoring model. The method was developed through a literature-informed design process and an expert-panel refinement stage. The paper describes factor definitions, scoring rules, weighting logic, and interpretation ranges, and positions Cx as a decision-support tool for managers and information systems governance. A compact illustrative comparison shows how Cx supports structured comparison and communication of assessment results. The poster positions Cx as a preliminary methodological artifact requiring empirical validation.

ABSTRACT. The increasing use of Automated Machine Learning (AutoML) in information systems development shortens development cycles and improves the short-term efficiency of analytical components, while delegating key design decisions to automated mechanisms. This paper argues that this practice may create a distinct form of technical debt: automation-induced technical debt. A lifecycle-oriented conceptual framework is proposed, identifying four dimensions of this debt: data, model opacity, governance, and integration, and its accumulation across successive phases of the system life cycle. The framework supports systematic debt management in AutoML projects and sustainable AI-driven systems.

ABSTRACT. AI-enabled information systems often operate on nonstationary data streams, where changes in data distribution may reduce the quality of automated decisions. This study presents a compact chunk-wise evaluation framework for comparing unsupervised and error-based concept drift detectors. The study focuses on two feature-space detectors, Walsh–Hadamard Drift Detector (WHDD) and One-Class Drift Detector (OCDD), and three error-based baselines: ADWIN, DDM, and EDDM. The detectors are evaluated on controlled synthetic streams with abrupt and gradual drifts. Besides the commonly used distance-based metrics D1 and D2, we include the count-consistency index R, which captures excessive or insufficient alarm generation. The results show that low D2 values alone may be misleading, because oversensitive detectors can cover true drift locations while producing unstable alarm patterns. WHDD provides a more balanced profile, combining competitive drift coverage with better alarm-count consistency. The study demonstrates that reliable detector assessment should jointly consider alarm localization, drift coverage, and alarm stability.

ABSTRACT. This work investigates the application of machine learning to demand forecasting in Supply Chain Management. Using the M5 Forecasting-Accuracy dataset, we develop a forecasting pipeline comprising a baseline and an enhanced feature-rich variant. A unified framework is used to train and evaluate diverse forecasting approaches, including linear models, neural networks, and Transformer-based architectures, in both univariate and multivariate settings. Results highlight the importance of data representation. On the baseline dataset, classical methods such as ARIMA achieve the best performance in univariate forecasting, while Transformers perform poorly. On the enhanced dataset, TimesFM 1.0 attains the lowest mean absolute error across all horizons, with linear models and shallow RNNs providing competitive, computationally efficient alternatives. Additional covariates generally offer limited benefits and, for Transformer-based models, often lead to substantial performance degradation.

ABSTRACT. Verification of scanned document authenticity is important for digital archiving, information security, and forensic analysis. This paper proposes a reconstruction error-based method for detecting structural anomalies in scanned documents. A convolutional autoencoder is trained on authentic document patches and used to generate reconstruction error maps. These maps are then classified by a CNN to distinguish the original scans from the reproduced samples. Experiments on a controlled dataset show that reconstruction error maps provide discriminative information for detecting artefacts introduced by document reproduction. The results should be interpreted as a proof of concept for controlled reproduction detection rather than as a complete solution for all real-world forgery scenarios.

ABSTRACT. Information Systems Development (ISD) increasingly occurs in socio-technical environments, where systems are developed and evaluated in real-life contexts involving diverse stakeholders. Living Labs provide a setting in which stakeholder engagement and experimentation can shape the evolution of systems during development. This research examines the SynAir-G project digital sentinel system as a case of user-driven ISD in a Living Lab context. Using an interpretive case research approach, the study analyses how iterative testing, field experimentation, and stakeholder interaction influenced the development of interconnected system components, including a mobile application, monitoring dashboard, and supporting services. The findings show that engagement cycles and real-life experimentation function as mechanisms through which system requirements, design decisions, and governance arrangements evolve. The paper discusses implications for organising ISD processes in multi-stakeholder real-world environments.

ABSTRACT. Quantum key distribution or QKD is entering early network implementation while integration with classical information systems remains difficult. This poster paper presents Holistic Quantum Integration as an ontology based requirements engineering framework for QKD based security systems. The study combines a PRISMA guided review of 48 studies, thematic analysis of 26 high quality studies, tradeoff analysis, and semi structured interviews with 26 practitioners. The resulting framework includes four requirement facets and a 47 item questionnaire for structured elicitation. The contribution addresses ISD researchers and practitioners integrating quantum security into organizational and technical environments.

ABSTRACT. Photovoltaic (PV) power forecasting is important for renewable-energy planning and decision support. Besides accuracy, interpretability is relevant because it helps explain how meteorological variables influence the forecasted PV output. This paper presents an interpretable neuro-fuzzy forecasting model based on feature selection, feature weighting, shared fuzzy sets, and population-based optimization. Input variables are selected using three complementary methods: correlation-based ranking, Random Forest importance, and an ANOVA-based univariate test. The resulting weights are incorporated into a Takagi-Sugeno-Kang Zero-Order system. The results show that the proposed method should be interpreted as an accuracy-interpretability trade-off: selected XGBoost baselines achieve slightly lower MSE values, while the neuro-fuzzy model provides competitive accuracy with explicit fuzzy rules and readable fuzzy sets.

ABSTRACT. Digital transformation strategies (DTS) in the micro and small enterprise (MSE) sector remain underexplored in the existing literature. Moreover, the level of digitalization within this sector is largely unknown, as European Commission reports provide data only for the broader small and medium-sized enterprise (SME) category. This paper presents the results of a literature review conducted using the Scopus database, focusing on DTS in the MSE sector. The findings of this review informed the design of a primary empirical study that the author intends to conduct among MSEs in Poland. The goal of the planned research is to assess the level of digital intensity across different MSE industries and to identify best practices in the implementation of DTS. The study will encompass key dimensions such as digital strategies, success factors, applied technologies, implementation barriers, and the impact of digital transformation on the functioning and performance of MSEs.

ABSTRACT. Digital transformation has become a critical success factor for organisations to survive in this highly-competitive and technology-driven world. However, the adoption and expansion of digital infrastructures have increased organisations’ exposure to cyber threats. Cybercrime is a complex and evolving phenomenon that requires collective effort across multiple stakeholders to effectively address it. The basic premise of this paper is that media reporting of cybercrime, and specifically its victim organisations, plays an important role in shaping organisational and societal responses to cybercrime. This research-in-progress work examines UK media reporting of cybercrime incidents involving organisational victims, with preliminary findings identifying seven emerging frames that capture the dominant narratives through which these incidents are portrayed. The ultimate aim of this work is to develop recommendations for constructive media reporting that strengthens cyber resilience and supports digital transformation.

ABSTRACT. In this paper, we propose a novel approach for the analysis of hierarchical multi-block data based on fuzzy techniques. The proposed method enables the identification of significant relationships within and between data blocks, while ensuring interpretability of the hierarchical multi-block structure. Empirical validation of the method on data from the ESG4PMChange and SPM² projects, assessing Sustainable Project Management and Environmental, Social, and Governance competencies, demonstrates the method's ability to uncover interpretable hierarchical dependencies that are robust to different aggregation strategies.

ABSTRACT. This study addresses the need for resilient critical‑infrastructure protection amid rapid digital transformation. While technological advances improve efficiency and connectivity, they also introduce complex cybersecurity and privacy risks. To overcome fragmented treatments of cybersecurity and cyberprivacy, we employ a multidisciplinary, interpretative approach grounded in socio‑technical systems theory to develop a unified framework synthesizing models, standards, and strategies across governance, risk analysis, socio‑technical factors, interdependencies, cyberprivacy and data governance, protection and mitigation, and evaluation and continuous improvement. Theoretically, it bridges technical controls and organizational behavior; practically, it offers a holistic basis for risk management and adaptable protection strategies. Although developed using the energy sector as a reference, the framework’s abstraction supports cross‑sector application.

ABSTRACT. Public administration processes often suffer from hidden bottlenecks and complex procedural heterogeneity. While traditional process mining discovers workflow mappings, it struggles to provide contextual prescriptive solutions. We propose a Neuro-Symbolic dual-pipeline architecture. The first pipeline utilizes predictive machine learning, applying Strict Temporal Splitting, to classify delayed cases without data leakage. The second introduces a Multi-Agent Large Language Model (LLM) ensemble utilizing Bottleneck-Anchored Context Windowing (BACW). By standardizing semantic milestones and deterministically bounding contexts within event logs, the system autonomously synthesizes evidence-based root-cause hypotheses. Evaluated on the BPIC 2015 dataset, this architecture bridges the gap between predictive monitoring and actionable process redesign, significantly reducing LLM hallucinations compared to full-trace baselines.

ABSTRACT. Rapid growth in AI research has increased the need to understand its development across Europe, particularly at the intersection of economics and finance. This paper identifies the main European knowledge hubs in this area and examines whether countries follow different patterns of thematic specialisation. Using OpenAlex metadata for 2000–2025, it analyses publications by researchers affiliated with European universities through fractional counting, citation-weighted indicators, and topic modelling, identifying 12 latent topics. The results reveal strong publication growth, a hierarchical but increasingly multipolar geography of knowledge, and national specialisations. Countries’ competitive positions depend not only on research output, but also on the internationalisation and network integration of their scientific systems.

ABSTRACT. Reliable IMU-UWB motion estimation depends on upstream analytical choices, not only on predictor design. We frame wearable motion estimation as a reference-aware problem involving UWB-derived reference construction, IMU sensor selection, and temporal context. The proposed workflow reconstructs and cleans 2D UWB trajectories, aligns them with IMU streams, and creates learning-ready inputs for supervised speed estimation. On 619 indoor sessions from 73 participants, these choices affected accuracy and robustness. The best subject-disjoint configuration achieved RMSE 0.401 and Pearson correlation 0.791; excluding magnetometer channels improved robustness under unseen conditions.

ABSTRACT. Educating the younger generation about AI capabilities and its responsible use is crucial. Still, the high entrance threshold of basic programming skills and the domain complexity is disen- couraging for both students and teacher. This paper presents a no-code concept for teaching AI through artificial empathy. It shows how reinforcement learning, multi-agent systems, neural networks, and artificial empathy can be introduced through a non-scripted, adaptive swarm of mobile robots. The robots communicate visually, allowing students to observe behaviors and dynamics real time, in classroom environment. The paper presents early results, a prototype platform, and an exemplary research scenario.

ABSTRACT. This text presents new research on the educational benefits of using a deeply immersive and highly engaging virtual escape room based on mathematical puzzles. It consists of a hall and three themed rooms containing 13 puzzles that develop mathematical thinking and was implemented on the Unity platform by three students of the authors for a CAVE-type installation in the Immersive 3D Visualization Lab located at the Gdańsk University of Technology. The new experiment involved inviting students to an escape room before they started their first year of studies and testing whether solving puzzles in the escape room resulted in higher grades in the Linear Algebra course. The students were divided into five-person groups, each of which had an escape room at their disposal for 90 min. Before and after playing in the escape room, they completed fettle-satisfaction surveys and knowledge tests. After the Linear Algebra exam, it was checked whether the results of the students participating in the experiment differed from the other students’ results.

ABSTRACT. People increasingly use Google and ChatGPT to obtain health information before consulting healthcare professionals. This study compares the two platforms across accessibility, clarity, trustworthiness, empathy, accuracy, and risk. It also introduces the TRUST framework (Traceability, Realism, Understandability, Safety Orientation, and Timeliness) for evaluating digital health information systems. The findings show that Google provides broad and verifiable information, while ChatGPT improves understanding through conversational interaction. The study highlights the potential of combining both approaches to support safer and more reliable digital healthcare.

ABSTRACT. The goal of this paper is to propose a set of metrics for evaluating the development of Big Data projects using Agile methods. The paper identifies a gap in the lack of integrated approaches that combine Agile practices with Big Data Analytics Systems, which are inherently data driven. Based on a survey conducted, several issues were identified, including the main factors contributing to project failure. These findings led to the development of a proposed three-layer framework with metrics for such projects.

ABSTRACT. In the context of the information society and the digital economy, cybersecurity is becoming an integral part of public administration. Given the rapid changes in technology, implementing appropriate procedures and recommended best practices in cybersecurity is particularly important for ensuring the security of public administration. This article aims to determine the impact of implementing security procedures, organizational standards, and systematically improving employee competencies on ensuring cybersecurity in public administration. The results suggest that a comprehensive cybersecurity management approach, incorporating organizational and educational elements, may be inadequate in the face of current threats.

ABSTRACT. An analysis of the strategies and measures of strategic success of IT companies, viewed from the perspective of their competitive position, allows us to identify the factors behind their success. Based on a study conducted in 2024 and 2025 among European IT companies and green ICT developers, respectively, we identify the dominant strategic approaches among formal, behavioral, and hybrid. We also highlight the dominant measures of strategic success. The results indicate that behavioral strategies dominate the IT sector, although companies also widely use formal and hybrid approaches. Product quality and profit remain the most used measures of strategic success. Companies with a strong market position focus on quality, market value, and international expansion.

ABSTRACT. Constructing models is a common practice in software engineering that establishes understanding between various actors involved in building a software system. To address dynamic behavior within the problem domain, process models are used. Various languages that support behavioral modeling exist. Each language defines different types of elements and concepts, which impairs model interoperability. In this paper, we focus on OntoUML and the Object-Process Methodology (OPM) and introduce a transformation of OntoUML models of events into OPM. We describe the rules of the transformation and perform a case study to validate it.

ABSTRACT. Digital financial inclusion has become an increasingly important issue in ageing societies, where access to financial products and services is increasingly provided through digital channels. This article examines the role of electronic banking in digital financial inclusion among older adults in Poland. The analysis is based on survey data collected in the fourth quarter of 2023 from 380 households of older adults. The study compares the use of financial products and services by electronic banking users and non-users. The findings indicate that electronic banking users demonstrate a higher level of activity in the financial services market, particularly in relation to online payments, payment cards, and insurance products. Some older adults, despite holding a bank account, do not use electronic banking, which may limit their participation in the digital financial market and increase the risk of digital financial exclusion. Electronic banking may act as a gateway technology, enabling broader use of digital financial services. From an information systems perspective, promoting technology adoption requires reducing barriers such as technology anxiety, perceived risk, lack of trust, and interface complexity. The study contributes to the literature on digital financial inclusion and provides practical implications for designing accessible, trustworthy, and senior-friendly digital financial services.

ABSTRACT. This study investigates machine learning approaches for toxicity prediction using the Tox21 dataset, which contains 11,760 chemical molecules represented as SMILES sequences and 12 binary toxicity prediction tasks. The dataset is characterized by class imbalance and missing labels, making toxicity prediction particularly challenging. Two approaches were evaluated: a single-task transformer-based model (ChemBERTa) and a multi-task multilayer perceptron (MLP) using Morgan fingerprints. The experiments were conducted using stratified 5-fold cross-validation, with ROC-AUC as the primary evaluation metric. The results indicate that the multi-task approach achieved higher predictive performance (ROC-AUC = 0.8623) than the single-task ChemBERTa model (ROC-AUC = 0.8488), while requiring substantially lower computational resources. Additionally, for the evaluated SR-HSE task, partial fine-tuning of ChemBERTa reduced the number of trainable parameters from 44.1M to 14.8M with only a moderate decrease in predictive performance.

ABSTRACT. This exploratory study investigates AI-related discourse in practitioner-oriented IT conferences. The analysis covers presentations from six conferences in Poland and the United States between 2018 and 2025. AI-related presentations were identified and classified into seven AI subdisciplines using a hybrid text classification approach. The results show a strong increase in AI-related topics after 2022, with higher levels of interest observed in the United States. Machine Learning emerged as the dominant subdiscipline and the central element of the AI ecosystem. The findings demonstrate that industry conferences can provide valuable insights into technological trends and the evolution of AI within professional communities.

ABSTRACT. The rapid development of artificial intelligence is having a significant impact on and transforming the way organisations operate on social media – particularly in terms of the significant technological, social and reputational risks it poses. This applies to the creation of content shared on social media, its moderation, and its consequences – namely, user behaviour. Current research on the use of AI in social media focuses on phenomena such as disinformation, deepfakes and algorithmic content personalisation. Despite the growing use of AI in digital communication, however, the issue of systemic risk management remains insufficiently researched. This exploratory article identifies key risk categories and proposes a conceptual model for managing these risks – the AISMRF (AI Social Media Risk Management Framework) – which can support Information Systems Development.

ABSTRACT. This article examines the double exclusion experienced by older adults: digital and energy-related within the context of inclusive information systems and smart environments. The study aims to assess the extent to which scientific literature conceptualizes the coexistence of these two forms of exclusion. An analysis of 302 publications in Scopus revealed that despite growing interest in both energy poverty and digital exclusion, the literature rarely addresses these phenomena in an integrated manner, particularly in relation to older adults and information system design. The findings highlight the need for interdisciplinary research and design approaches that combine technological accessibility with users' material living conditions, enabling a more realistic and contextual understanding of inclusivity in information systems.

ABSTRACT. Design thinking enables collaborative innovation by supporting stakeholders in externalizing knowledge through haptic artifacts. However, these artifacts remain informal and are difficult to integrate into conceptual modeling and AI-supported information systems development. This paper positions Scene2Model as a platform for digital design thinking, in which multimodal inputs are transformed into adaptive conceptual models. These models represent digital twins of design thinking that evolve from informal artifacts into processable knowledge representations. In contrast to predefined modeling approaches, Scene2Model supports the adaptation of modeling constructs during runtime. By utilizing conceptual models for information retrieval and AI-based processing, they act as an interface between human-centered design and machine-processable knowledge. The platform perspective further enables the reuse and integration of modeling capabilities across application contexts. Illustrative cases demonstrate the potential for linking co-creative design with model-driven and AI-driven information systems development.

ABSTRACT. This paper presents the design, implementation and validation of the RACI matrix as an artefact supporting the management of responsibilities in a matrix structure in an international delivery organization. The study was conducted according to the Design Science Research (DSR) approach, which aims to create useful and scientifically grounded solutions. The RACI matrix was implemented in three project portfolios in 2023 and validated and iteratively refined in 2024. The study contributes to both the theory and practice of project management by providing a validated artefact that can be applied to complex organisational structures.

ABSTRACT. In the era of intensive development of artificial intelligence, its methods are increasingly used to solve complex business problems. One such area is the implementation of ERP systems, which are characterized by high variability and the risk of budget overruns. The aim of this paper is to investigate whether AI methods can be used to forecast deviations between planned and actual budget values. To this end, three predictive models are evaluated based on real-world data from 77 implementations carried out over a 10-year period by a company ERP Serwis sp. z o.o. sp. k. The performance of the models was evaluated using standard regression measures and interpretability analysis. The results indicate a limited but meaningful predictive ability of machine learning models to identify the risk of budget overruns.

ABSTRACT. Accurate differentiation between cognitively unimpaired individuals, mild cognitive impairment (MCI), and dementia remains challenging, particularly when heterogeneous data must be integrated within a single predictive framework. This study proposes a strategy-aware multimodal late-fusion approach for three-class classification using 2D MRI, 3D MRI, and tabular clinical data, explicitly accounting for different diagnostic priorities at the fusion stage. The best results were obtained by a balanced late-fusion configuration based on ResNet-18 for both imaging branches and Random Forest for tabular classification and fusion, achieving 0.96 accuracy, 0.97 macro-precision, 0.96 macro-recall, and 0.96 macro-F1, with strong recognition of the clinically challenging MCI class.

ABSTRACT. Information systems project management (ISPM) research is fragmented across diverse topics, methods, and publication venues, making it difficult to assess its development and identify research gaps. This paper proposes an analytical framework for characterizing ISPM research through three key dimensions: bibliometric, research techniques, and project-management-specific aspects. The framework supports the systematic analysis of research corpora across national, institutional, and thematic contexts. Its application is illustrated using a corpus of publications authored by researchers affiliated with Portuguese organizations. The results highlight a predominance of technology-oriented and project-level studies, frequent use of surveys and case studies, and a strong interest in PM tools, agile and hybrid approaches, and the evaluation of project success. The framework provides a basis for research assessment at different levels and for future agenda-setting.

ABSTRACT. Research on the credibility of responses to online surveys has thus far omitted geo-questionnaires, which are web-based surveys which involve participants with an interactive map. This paper proposes a heuristic credibility assessment model for online geo-questionnaires, based on ten behavioral indicators derived from paradata, including response time, zoom interactions, inactivity periods, and geometric consistency of drawn routes. The model was implemented in an online geo-survey platform and evaluated on data collected from 44 university students, who completed a geo-questionnaire about the city of Gda´nsk with engagement control questions. Results show that the proposed indicators effectively distinguish between low-knowledge and low-engagement respondents, and that credibility-based filtering can increase average respondent scores by up to 12.1%. The findings confirm that paradata-driven credibility assessment can serve as a practical quality filtering mechanism in geo-questionnaire analysis.

ABSTRACT. Crowdfunding platforms are usually discussed as alternative financial mechanisms. However, from an information systems (IS) perspective, they can also be understood as digital ecosystems that integrate users, data flows, platform architectures, and regulatory governance. This paper proposes a preliminary conceptual reframing of European crowdfunding platforms in terms of digital financial ecosystems rather than purely financial intermediaries. Drawing on selected IS literature and institutional reports from ESMA, OECD, and the Cambridge Centre for Alternative Finance, the paper develops a dual financial/IS interpretation of crowdfunding platforms. The main contribution is a comparative framework that translates financial characteristics of crowdfunding into IS concepts such as data flows, network nodes, distributed participation, system interactions, and governance layers. The paper also outlines a path for future empirical research aimed at validating and extending this conceptual framework through document analysis and semi-structured interviews with platform operators, regulators, and other ecosystem actors.

ABSTRACT. Through the infrastructure of our partner Next Mobile P.S.A. around 150-170 million unfiltered connection attempts are transferred every month. The data of 3.3 million phone numbers from August 2024 was divided into 21 classes. Then, the research used several machine learning and artificial intelligence methods, e.g. k-means, LSTM, SVM, decision tree (Fine Tree), optimized decision tree, Gaussian Naive Bayes, KNN, and Random Forest. The best approach was to use in the first step unsupervised methods, then compare the numbers with the online opinion even if it is limited to around 200 thousand numbers, and use supervised methods at the end with classes as cluster numbers. We have received the accuracy around 99.4% with true positive and true negative rates 99% as well for the chosen group numbers. Our final solution, which is a blend of machine learning and statistical analysis, is the final tool to spot suspicious numbers.

ABSTRACT. Mobile applications support travel planning, navigation, and access to urban transport services. Within the Society 5.0 paradigm, such technologies should improve quality of life through user-centered solutions. However, limited evidence is available on how cyclists with different levels of cycling activity use navigation, ticketing, and city applications. The study examines mobile application use among regular cyclists, occasional cyclists, and non-cyclists in Poland. The findings show that more than half of the respondents use a bicycle in urban mobility and that application use varies according to cycling activity and application type. Navigation applications are used most frequently, whereas more general city applications are used less often. The findings suggest that the adoption of digital mobility tools is associated with their practical relevance and ability to support users during everyday travel.

ABSTRACT. The objective of this study is to identify sectoral adoption patterns and value proposition drivers among institutional decision-makers, and to propose an empirically grounded business model configuration for UAV delivery services in medium-sized cities operating within the European U-Space framework. Drawing on a CATI survey of 183 institutional entities in Stargard, Poland (approx. 67,000 inhabitants), we analyze adoption readiness across NACE sectors and differentiating factors between interested and uninterested entities. Safety perception and expected benefits emerge as the primary adoption drivers, with the highest interest rates in health and social work (36.36%), transport and storage (30.00%), and manufacturing (25.00%). We propose a five-component business model configuration integrating empirical demand patterns with U-Space operational requirements and discuss implications for digital platform ecosystems within U-Space infrastructure.

ABSTRACT. Modern data centres constitute the physical substrate of modern information systems (IS), yet their operational management often relies on fragile, manual workflows. This poster presents a hybrid NetDevOps automation framework designed to resolve the structural vulnerabilities of purely stateless Infrastructure as Code (IaC) pipelines. The architecture introduces a model-driven mathematical schema that deterministically derives expansive routing identifiers (e.g., 4-byte ASN, loopbacks, and 24-bit VNIs) directly from topological semantics. Simultaneously, it integrates NetBox as a stateful Single Source of Truth (SSoT) exclusively for constrained resources (e.g., VLANs), effectively preventing index-shift anomalies. The orchestration layer combines Ansible playbooks, Jinja2 templates, and the NETCONF protocol to enable transactional, idempotent provisioning of Juniper network devices. Validated in a dual-site EVPN-VXLAN emulation environment, the framework acts as a strict procedural enforcer, eliminating configuration drift and elevating network management into a reproducible, declarative, and auditable engineering discipline.

ABSTRACT. This work compares two approaches to multi-tenant isolation and load balancing on a single-site, IPv6-only testbed: (1) F5 BIG-IP LTM with routing domains and administrative partitions and (2) HAProxy Community with Keepalived and Linux VRF. They are assessed across data plane isolation, management plane isolation, high availability, and operational characteristics; load-balancer takeover was measurable only for the open-source stack. Both approaches blocked all tested cross-tenant traffic through separate routing contexts and showed similar behaviour after pool-member failure. Their main difference is the management plane: F5 provides native partition-scoped administration, whereas the tested open-source stack requires additional controls.

ABSTRACT. Large language models (LLMs) are increasingly used as on-demand assistants for software tasks, yet their educational value under everyday, unguided use remains uncertain. This study tests whether unrestricted LLM access improves beginners’ code comprehension. In a controlled classroom experiment, undergraduate novices practiced comprehension tasks in Python, SQL, and HTML/CSS either with unrestricted access to a public LLM or without AI support. Matched pre- and post-tests measured learning gains, with LLM access disabled during testing. Both groups improved modestly; however, mean gains did not differ significantly between conditions across languages, and effect sizes were small. The results indicate that LLM availability alone is insufficient to measurably improve novice code comprehension without instructional scaffolding, and they motivate governance and training approaches for sustainable AI-assisted information systems development.

ABSTRACT. The utilisation of artificial intelligence (AI) is progressively becoming imperative for port managers. However, the extent of its implementation remains challenging to ascertain due to the absence of a universally adopted instrument within extant literature. The objective of the present article is to address the extant research gap by proposing a composite indicator with the capacity to determine the extent of AI implementation in ports. The conceptual framework of the article is predicated on an analysis of extant research, which indicates that the measurement of the degree of AI adoption should be a complex process, i.e. it should encompass multiple elements. The AI Adoption Index for Seaports (AAIS), as outlined in the article, has the potential to serve as a foundation for comparative analyses and the identification of implementation gaps in seaports.

ABSTRACT. This paper presents a fuzzy relational approach to discrete-time survival analysis based on fuzzy relational inequalities for modeling and predicting hazard probabilities over time. The proposed method combines data fuzzification with interval-specific fuzzy relational models, providing both accurate prediction and an interpretable description of the relationships between predictors and the risk of failure. Three fuzzification methods (sigmoid, Gaussian, and Gaussian smoothing) and four t-norms (minimum, product, Lukasiewicz, and Fodor) were investigated. The proposed approach was evaluated on three benchmark datasets. The experimental results demonstrate that both the fuzzification method and the choice of t-norm affect prediction performance, with Gaussian smoothing and the Lukasiewicz t-norm providing the best overall results. The proposed framework offers a flexible and interpretable tool for discrete-time survival analysis, particularly for datasets characterized by uncertainty, gradual transitions, or nonlinear relationships.

ABSTRACT. This paper aims to analyse the interpretability of public transport delay predictions using post-hoc model interpretability methods. Using Gradient Boosted Machines and the Shapley Additive Explanations (SHAP) method, we identify the factors with the greatest impact on predicted delays. The results were presented as a list of the most important factors contributing to the delay, along with a spatial analysis of the results for individual public transport lines and the entire urban area. The tests were performed on monthly data from streamed localisations of public transportation means in the City of Warsaw, amounting to more than 66 million records. The proposed solution increases the transparency of predictive models and can support public transport operators in making operational and planning decisions.

ABSTRACT. Clinical Decision Support Systems (CDSS) increasingly rely on AI components to support diagnostic processes. However, existing studies primarily focus on predictive performance, while lifecycle management, reproducibility, and operational aspects of AI deployment remain less explored. This paper proposes an MLOps-based framework for managing and evaluating AI-enabled decision services within clinical information systems. The framework integrates model versioning, experiment tracking, monitoring, and lifecycle management. As a proof-of-concept implementation, a compact Vision Transformer (ViT-like) architecture was applied to brain MRI classification. Experimental results demonstrate high predictive performance (99.31\% accuracy), stable training behavior, and operational feasibility of the proposed framework. The findings highlight the importance of evaluating AI components not only in terms of accuracy but also with respect to reproducibility, traceability, and operational feasibility.

ABSTRACT. Medical anomaly detection is challenged by subtle pathologies, limited abnormal training samples, and modality heterogeneity. Reconstruction-based autoencoders are widely used in anomaly detection, but may fail when anomalous inputs are reconstructed too accurately. In this paper, we propose a dual-criterion anomaly detection approach that combines image-space reconstruction evidence with latent-space deviation, using Adversarial Autoencoders. The approach was validated on seven heterogeneous benchmark datasets covering 2D and 3D medical images. Comparative evaluation against alternative autoencoder variants showed that AAE achieved the strongest average performance, with an aggregate score of 0.723. These results support the use of adversarially regularized autoencoders as a suitable foundation for dual-criterion anomaly detection in heterogeneous medical images.

ABSTRACT. Missing observations in air quality sensor networks can reduce the reliability of environmental monitoring systems. This paper proposes GCN-KAN, a hybrid model combining graph convolutional networks and Kolmogorov–Arnold Networks for PM data imputation by exploiting spatial and nonlinear temporal dependencies. Experiments conducted on multiple missing-data scenarios demonstrate improved reconstruction accuracy and robustness compared with baseline methods.

ABSTRACT. This study proposes a Geographic Information System (GIS)-based approach for assessing accessibility to urban facilities in the context of the “15-Minute City” concept. The study was related to the case of Szczecin (Poland). GIS network analysis was employed to generate 15-minute isochrones, enabling assessment of spatial accessibility to facilities across the study area. The analysis covered twelve types of facilities, grouped into three main categories: administration and retail, education, health. The results indicated a high concentration of facilities in the two central districts (on the left and right banks of the city), with lower densities in the outskirts. The findings also reveal discrepancies between the quantity and the diversity of services. Furthermore, the study demonstrates how GIS-based analytics can support the digital transformation of urban governance by providing data-driven accessibility assessments and spatial decision support.

ABSTRACT. Large language models (LLMs) are increasingly used to detect and evaluate sentiment in unstructured text. Configuring an LLM annotation pipeline requires researchers to make numerous design choices, raising the question of how sensitive annotation outputs are to these researcher degrees of freedom. This study investigates the reliability and validity of LLM annotations by systematically varying five pipeline parameters across four open-weight models, 150 multilingual hotel reviews, and six service quality attributes, producing over 61,000 annotation data points. Results show that annotations are highly reliable and largely insensitive to pipeline configuration, with internal consistency well above acceptability thresholds (Krippendorff’s α =0.839–0.944). Criterion validity is also high, with LLM composite scores correlating with external ratings at ρ = 0.816–0.874, significantly outperforming both VADER and multilingual BERT baselines.

ABSTRACT. A controlled laboratory environment is proposed for evaluating classical and machine-learning-based forecasting algorithms. The paper argues that conclusions drawn from real-world data may be systematically biased because the true data-generating structure is unknown. The study therefore constructs 15,768 reference time series with controlled length, noise, trend, seasonality and integer-valued structure. Eight forecasting methods are compared using RMSE and within-series rankings. Linear regression and ETS achieve the best overall accuracy, while Random Forest and XGBoost are competitive only in specific structural settings. The results show that algorithm selection should depend on identified demand characteristics rather than on aggregate benchmark accuracy alone.

ABSTRACT. This paper presents the design and evaluation of a maintainable hybrid generative architecture for automated music harmony generation from melody. The proposed system combines quantum-inspired candidate exploration over overlapping melodic contexts with explicit rule-based optimization to balance generative flexibility and structural control. The architecture is evaluated using explicit and reproducible metrics covering structural coherence, functional agreement, harmonic similarity, and robustness. The results show that the proposed approach produces harmonizations that preserve tonal structure and cadential behavior while allowing multiple valid harmonic realizations. Furthermore, the optimization layer improves structural coherence, stability, and predictability without requiring a training corpus. The study demonstrates that transparent and controllable hybrid generative systems can be systematically designed and evaluated within the context of Information Systems Development.

ABSTRACT. Training engineering students in correct machine operation procedures is challenging due to safety constraints, limited access to laboratory equipment, and the high cost of operating the equipment, especially in the event of user errors. While virtual reality (VR) is increasingly used in engineering education, many VR training systems emphasize open-ended interaction and visualization, offering limited support for enforcing procedural correctness. The paper presents a VR simulator for closed-ended procedural training of milling machine operation, developed for a CAVE-type (Cave Automatic Virtual Environment) installation. The simulator integrates a virtual milling machine, CNC (Computer Numerical Control) control terminal, tooling components, and instructional elements. Procedural correctness is enforced by dynamically constraining user interactions according to the active training scenario. A representative machine preparation scenario was implemented and evaluated through a user study conducted with eight students in the CAVE environment. The results indicate that the proposed approach is usable and perceived as beneficial for supporting structured procedural training in an immersive educational setting.

ABSTRACT. This study investigates how different color combinations affect individuals under simulated visual impairments using virtual reality (VR). The experiment involved participants with normal vision as well as participants with minor visual disorders; however, all participants were treated uniformly and experienced simulated visual impairments within the CAVE-type (Cave Automatic Virtual Environment) installation. Participants performed everyday tasks under various color combinations, while task completion time was measured automatically from task initiation to completion. The study was conducted for architects to help them understand how different color combinations can affect the perception of the environment and the performance of everyday tasks by people with visual impairments. The results indicate that most of the color combinations do not provide measurable benefits (black-blue, black-yellow, green-white, redyellow), while the red-white color combination possibly retard task performance.

ABSTRACT. Identifying a fixation is a crucial step of an eye-tracking pipeline, especially for scenarios where a user might need to operate a device hands-free, either due to mobility impairments or because they are in a situation where manual operation is not feasible. However, distinguishing between a prolonged fixation and a random eye movement is often done arbitrarily, with fixed thresholds applied to the gaze points. Our work proposes a user-tailored fixation identification method, calculating the gaze's dispersion and velocity to effectively discriminate between fixations and saccades. To do so, a solid gaze estimation is needed: we fine tuned a multi-Mobile Vision Transformer model on the GazeCapture dataset, obtaining lower errors than the reference iTracker model. Finally, we adapted the standard 13-points calibration procedure to incorporate the computation of the dispersion and velocity of the gaze points based on the I-VT and I-DT algorithms, obtaining calibration errors as low as 0.51±0.12 cm.

ABSTRACT. Requirements elicitation is a resource-intensive task that artificial intelligence (AI) can potentially streamline. This paper aims to compare large language model (LLM)-generated aggregated response distributions with requirements from human participants for an app designed to enhance creativity during training sessions. The exploratory research is based on: a questionnaire administered to human respondents, results from the same survey across LLMs - GPT, Gemini, Grok, Mistral, Claude; and a comparative analysis of results. The results show that LLMs can mimic preferences for minimalistic and intuitive interfaces. However, for more detailed aspects, differences emerge between human responses and LLM-generated aggregated response distributions. This study contributes to practice by demonstrating that LLMs can support the identification of general user preferences, but should not replace human users due to systematic biases.

ABSTRACT. The aim of this paper is to propose a framework addressing Big Data quality aspects in the context of Web data. The identified research gap, based on a review of existing literature, is that Big Data quality is typically evaluated using broad and general dimensions, such as data consistency, without providing detailed methodologies for measurement or specifying metrics suitable for continuous quality monitoring. This paper focuses on challenges identified in case studies involving the use of Web data, particularly in the context of large-scale web scraping, to derive information about enterprise characteristics.

ABSTRACT. Acquired color vision deficiencies may provide useful functional information in monitoring of various diseases. This paper presents a revised, more conservative architecture and baseline evaluation of the Interactive Detection of Color Vision Disorders (idCVD) system. While the current implementation of idCVD focuses on generating randomized color-pair decisions along protan, deutan, and tritan confusion axes and on recording both response correctness and time, in this paper a Bayesian adaptive layer is proposed for selecting subsequent stimuli according to expected information gain. This layer is treated as a design target rather than an empirically validated component. The evaluation covers 317 completed studies from 287 participants, producing 39,447 decision events. The observed trial-level accuracy was 97.79% with rare timeout responses (0.19%). Response time captured information not reducible to accuracy alone, proving that adding a Bayesian layer to the diagnostic methos forms a solid research agenda for clinically validated adaptive self-monitoring of color vision.

ABSTRACT. This paper addresses three closely related optimization problems: the set cover problem, the decision rule length minimization problem, and the maximum rule length minimization problem within a complete system of decision rules. New algorithms for the set cover problem are proposed and applied to the decision rules domain.

ABSTRACT. Generative artificial intelligence, particularly large language models (LLMs), offers significant potential for improving access to legal information and supporting decision-making. However, integrating these technologies into legal information systems introduces challenges related to reliability, explainability, jurisdictional constraints, and the temporal validity of legal knowledge. This paper examines these challenges and how they shape the architecture and functionality of legal information systems. To address them, we introduce JurisGraph, a conceptual model that captures structured legal knowledge. By combining structured representations of legal concepts and relationships with LLMs and retrieval-augmented generation, supported by multi-agent orchestration, such systems can improve the transparency, consistency, and trustworthiness of their outputs. The paper highlights the importance of combining unstructured generative capabilities with structured domain knowledge in the design of reliable AI-driven legal systems.

ABSTRACT. This work analyzes the vulnerability of Large Language Models (LLMs) deployed in voice-based conversational systems. While most prior research focuses on text-based interfaces, less attention has been given to voice-based assistants, where user input is transcribed from speech and responses must conform to structured output constraints. We compare a simple, intuitively written prompt with a prompt engineered according to best practices identified in prior research. We also evaluate the impact of incorporating additional behavioral policies directly into the prompt. Our results show that increasing prompt template complexity significantly improves structured-output compliance and reduces injection attack success rates. For language-forcing attacks, the attack's success rate decreased from 56.7% for a simple prompt without policies to 14.2% for a complex prompt with embedded policies. However, some format-override attacks remain highly successful, highlighting the limitations of prompt-only defense mechanisms.

ABSTRACT. In recent years, Artificial Intelligence (AI) has advanced rapidly, becoming a routine part of people's lives worldwide. Despite its many benefits, it has also been associated with negative incidents. In response, governments, international organizations, and companies have released documents outlining principles or guidelines for ethical AI. However, the literature also reports ongoing challenges in their practical implementation. In parallel, maturity models have also been developed to assess and guide the implementation of ethical requirements for the development and use of AI systems. This study presents a systematic literature review to understand the state of the art in maturity models focused on ethics in AI and to identify remaining gaps. We found and analyzed 10 such maturity models and concluded that none were sufficiently validated or comprehensive. These findings point to the need for further research on maturity models focused on AI Ethics.

ABSTRACT. Large language models (LLMs) are increasingly deployed in interactive settings where users provide feedback in natural language, often with an emotional component. This paper examines whether such feedback is associated with systematic changes in LLM outputs. Using 42 Polish prompts across six everyday domains and 20 additional feedback prompts, we compare responses from GPT-4o mini, Mistral Small 24B Instruct, and Phi-4 before and after emotionally valenced contextual prompting. We analyze response length, type-token ratio, syllables per word, and Jaccard similarity. A survey of 50 participants evaluates perceived changes in tone, structure, and usefulness. Results provide preliminary evidence that emotionally valenced feedback modulates verbosity and lexical diversity, and that human raters prefer output produced after feedback. We interpret these findings as behavioral modulation driven by context rather than adaptation at the parameter level and discuss implications for information systems design.

ABSTRACT. In the light of the growing popularity of the freelance market and numerous challenges posed by this fast-changing market, the current study aims to identify freelancers’ skills required on the market. In so doing, the study builds on the preliminary research conducted in multiple countries: Indonesia, Poland, Romania, Serbia, Spain, and Ukraine. During the study, semi-structured interviews have been conducted with professionals and companies operating on the freelancer market. The data analysis allowed us to identify the most important factors and areas for freelancers that require improvement. The analysis also allowed us to gain insight into the role of culture and socioeconomic development. The main findings illustrate the paramount role of freelancer’s personal attributes such as self-organization and assertiveness, and also suggest the growing importance of soft skills relative to the level of the country’s socioeconomic development.

ABSTRACT. Robotics is currently developing dynamically, bringing the first commercial humanoid units to the market. One of the key functionalities of humanoid robots is verbal communication, resulting from their intended role in human interaction. The aim of this work is to design and evaluate the feasibility of implementing a local speech interface for a humanoid robot, operating without cloud services, with consideration of the target NVIDIA Jetson Orin platform. The paper presents a system based on open-source large language models, whose textual responses are converted into speech using built-in audio devices. The work focuses on implementation and validation in a test environment as a preparatory stage for deployment on the robot. The results confirm the feasibility of a local speech processing pipeline in a resource-constrained environment. The proposed solution serves as an educational platform for embedded AI systems.

ABSTRACT. Telemedicine systems rely on AI-based Automatic Speech Recogniton (ASR) modules for clinical documentation, yet audio codec selection, a frequently overlooked infrastructure decision, can silently degrade transcription accuracy. Framed as a design science study, this work introduces a codec admissibility framework: a prescriptive Information Systems (IS) artefact that classifies audio codecs by their measured impact on Whisper large-v3 transcription of Polish medical speech. Using 2,000 annotated recordings across seven medical domains, we computed bootstrapped 95% confidence intervals for ΔWord Error Rate (WER) under three normalization strategies. AAC, MP3, Opus, and ADPCM cause no statistically significant degradation, while G.723, Speex, and the neural codec produce significant increases in ΔWER of up to 0.230 (95% bootstrapped Confidence Intervals (CI): [0.181, 0.230]) The resulting codec selection decision table provides concrete non-functional requirements for IS architects designing ASR-integrated telemedicine platforms.

ABSTRACT. The primary objective of the study was to analyse the connections between projects carried out under the three European Union framework programmes and concepts in the field of artificial intelligence. To address this problem, a method for identifying artificial intelligence concepts in project descriptions was developed. This method considers all concepts from the “Computer Science Ontology” that are related to artificial intelligence. The correspondence between AI concepts and project descriptions was determined using semantic similarity measures. The analysis covered all projects implemented between 2007 and 2025 as parts of FP7, Horizon 2020 and Horizon Europe frameworks (more than 80,000 projects). All analyses were conducted by the authors using Python programmes developed specifically for this study.

ABSTRACT. Practical command-line skills are essential in computing education, yet the ability to operate in AI-restricted, terminal-based environments remains underdeveloped by traditional teaching methods. This paper presents CLI2me, a command-line, text-based, and distraction-free educational system designed to promote documentation-driven, AI-assisted learning through structured, hands-on exercises in an isolated environment. Trainees solve tasks by consulting on-host manual pages within a personal virtual machine, and submit answers for automated evaluation inside a namespace-based sandbox. An AI feedback mechanism provides contextual explanations upon task completion, without interrupting the learning flow or requiring the trainee to leave the system. The system was deployed in a controlled academic environment, giving trainees full administrative access to their virtual machines, while the CLI2me session enforces a controlled, restricted and distraction-free environment.

ABSTRACT. As Network Intrusion Detection Systems (IDS) increasingly adopt unsupervised paradigms to detect zero-day threats, their vulnerability to adversarial evasion remains a critical security gap. We evaluate the robustness of three unsupervised models against a domain-restricted Masked FGSM attack, together with their inference latency. By enforcing categorical feature masks and discrete integer rounding, we craft protocol-valid adversarial samples. On UNSW-NB15 and CIC-IDS2017 all models prove susceptible, but the adversarially-learned latent space of the AnoGAN family can be more resilient than reconstruction-based Autoencoders; for the fast f-AnoGAN this advantage is strongly initialization-dependent, despite its inference time matching the Autoencoder, which was the most performant. These preliminary results suggest f-AnoGAN is a promising but unstable candidate for Network Intrusion Detection Systems and Next-Generation Firewalls.

ABSTRACT. This study explores the stakeholders and their roles in ICT-supported active and healthy ageing (AHA) from the perspective of seniors. The study builds on 60 in-depth interviews with seniors in Poland and Sweden and employs analysis considering differences in socioeconomic context. Our preliminary findings revealed seventeen stakeholders that were categorized into four domains related to social network, care system, information technology, and institutions and organizations. Compared with prior research, the study provides a more nuanced understanding of the AHA ecosystem. Our findings also suggest that as socioeconomic development increases, the role and awareness of technology might grow, while the importance of people remains unchangeably high.

ABSTRACT. This paper presents a preliminary GIS-based analysis of public transport accessibility with a focus on older adults in a mid-sized European city, using Greifswald as a case study. The study applies network-based methods, including walking-time and walking-distance isochrones, to assess access to bus stops in relation to the spatial distribution of residents aged 65 and over. Results indicate generally good coverage, with most residential areas accessible within a 5-minute walk or 300–600 meters distance. However, specific gaps in service provision were identified, particularly in peripheral seaside areas. The findings highlight the usefulness of GIS tools for identifying accessibility inequalities and supporting data-driven improvements in public transport planning for aging populations.

ABSTRACT. The dynamic development of generative artificial intelligence (Gen AI) is increasingly influencing information systems and user interactions with digital technologies. However, despite the growing availability of these tools, the extent of their practical use and associated risk perception remains unclear. This paper aims to analyse the adoption and use of generative artificial intelligence, as well as perceived risks, based on a quantitative survey of 429 participants at a regional job fair in the West Pomeranian Voivodeship, Poland. The findings indicate a high level of awareness of Gen AI, accompanied by a clear difference between private and professional use. While it is widely used in everyday activities, its application in professional environments remains limited. At the same time, most respondents perceive risks related to its development and express a need for further education. From an information systems perspective, the results highlight the importance of aligning AI-based solutions with users’ knowledge, trust, and risk perception, as these factors may influence their further development and implementation.

ABSTRACT. Hyperparameter optimization (HPO) for tree-based machine learning models remains computationally expensive due to exhaustive search and repeated cross-validation. This paper proposes a parallel e-fold Halving Grid Search (e-fold HGS) framework that integrates successive halving, adaptive e-fold cross-validation, and parallel execution to reduce evaluation cost. Experiments conducted with Random Forest and XGboost on six benchmark datasets show that e-fold HGS achieves statistically equivalent predictive performance to exhaustive Grid Search and original HGS, with all performance differences falling within overlapping 95% confidence intervals. At the same time, e-fold HGS delivers a substantial reduction in average execution time, achieving up to 37.5% and 46.64% in Random Forest and XGboost, respectively, with statistically significant improvements observed across the six datasets. These results demonstrate that e-fold HGS provides an efficient alternative for hyperparameter optimization without compromising predictive accuracy.

ABSTRACT. This study analyses the potential applications of artificial intelligence in supporting variable cost budgeting in small and medium-sized enterprises using spreadsheet-based systems. An exploratory experimental approach was adopted in the study to assess whether an AI tool can derive mathematical functions that accurately reflect the relationship between variable costs and quantity, considering various patterns of cost variability. The results of the study indicate that the quadratic functions generated by AI exhibit varying degree of fit, suggesting their potential usefulness in budgeting processes. AI has the potential to reduce the labour intensity and improve the responsiveness of budgeting; however, its results require careful validation prior to implementation.

ABSTRACT. The dynamic nature of skill requirements poses challenges for educators, job seekers, policymakers, and employers. Advancements in AI hinder the timely tracking of skill needs and the forecasting of future trends. This study analyzes the evolution of skill demands using job advertisements from 2021 onward for data science, analytics, and AI roles in Poland. For AI roles, the findings indicate a rapid rise in LLM and Generative AI skills, along with a shift from model-centric work to system-level AI integration, architectures, APIs, and workflows within organizational information systems. Across data and analytics professions, the persistence of core competencies such as Python and data infrastructure suggests that foundational technical skills remain stable, with AI complementing existing capabilities. The decreasing emphasis on GDPR, security, and data quality in some professions raises concerns. The study provides insights and skill-frequency data to support curriculum development, workforce planning, and future research on AI-driven labor-market transformation.

ABSTRACT. The abstract summarizes a Proof of Concept (PoC) of measuring the impact of stealth data poisoning with a fixed magnitude on a federated machine learning model. The model is created in Tensorflow Lite and is used to predict the strength of the mobile network. It operates on four physical mobile devices running the Android operating system, each representing a different mobile network. The experiment demonstrates the hardware and software setup (including limitations applied), initial results of accuracy and loss measurements, and indicates potential improvements for the future.

ABSTRACT. Earlier pilot work on Scrum teams using AI-assisted HCI design reported that generative co-design tools compress sprint feedback loops but introduce a hidden cost “AI process debt” that slows later iterations when model opacity is left unmanaged. That study proposed two lightweight practices (a daily model-state check and an explainability-extended Definition of Done) without formalising them. This poster offers a conceptual extension rather than new empirical data. Recent quantitative work on process debt in agile teams by Gustavsson et al. provides the missing empirical anchor, and the rise of agentic AI coding assistants amplifies rather than replaces the opacity problem identified for generative prototyping. We consolidate both safeguards into a single framework, AID-Scrum (AI-Debt-aware Scrum), built from four interlocking artefacts: an AI Process Debt Ledger, a model-state traffic-light protocol for Daily Scrum, an XAI-extended Definition of Done, and a five-level maturity model. We close with five falsifiable propositions to guide empirical validation. The contribution is a research agenda and a set of ready-to-pilot artefacts for Scrum teams moving from generative co-design to agentic co-delivery.

ABSTRACT. This paper studies sinusoidal modulation in rough inclusion classifiers. Attribute contributions are controlled by three interpretable parameters: amplitude, period, and vertical shift. The revised formulation normalizes the sinusoidal phase with respect to the number of attributes, making the period parameter directly interpretable. Experiments on five benchmark datasets with MCCV-10 validation show that random sinusoidal perturbations usually preserve baseline behavior, while targeted tuning improves balanced accuracy.