Production CRM systems increasingly use large language models, yet typical Retrieval-Augmented Generation (RAG) implementations suffer from knowledge staleness due to 5–10 min batch processing cycles. This paper presents a streaming RAG architecture for business CRM applications that provides real-time knowledge updates with average document-to-query propagation latency of 3.1 s and strong retrieval quality. The event-driven system uses Apache Kafka for document ingestion, Rust microservices for embedding generation, PostgreSQL with pgvector for vector storage, and GPT-4 for response generation. On 62 insurance policy documents from 20 users and 102 test queries, mean document-to-query propagation latency was $3.1 ~\mathrm{s}, 75-150 \times$ faster than batch processing, with retrieval quality metrics of Precision@5 = 0.398, MRR $=0.938$, and NDCG${@} 10=0.942$ consistent with values reported in prior literature. Additional load testing with simulated users verified production-grade performance stability (P95 latency $<10.33 ~\mathrm{s}$), suggesting that streaming designs may mitigate the knowledge-currency vs. system performance trade-off in production CRM applications.

This paper benchmarks LLM-generated synthetic data for fine-tuning RoBERTa-base on two GLUE tasks (SST2 sentiment classification and MRPC paraphrase detection) under a low-resource setting with 1,000 real training examples per task. Real-only, synthetic-only, and hybrid (1 k real + 1 k synthetic) regimes are compared using data from eleven contemporary LLMs. Results show that synthetic-only training remains below real-only baselines, but hybrid training consistently improves performance: on SST-2, the best hybrid configuration nearly matches doubling the real data, while on MRPC gains are smaller but positive. LLM-generated text is most effective as a supplement rather than a replacement for human-labeled data.

This paper presents the design and implementation of a prototype chatbot system based on the Retrieval-Augmented Generation (RAG) architecture, applied in a scientific research institute to improve knowledge access. The system combines semantic search over a vector knowledge base with response generation using large language models, enabling contextually relevant institutional information. A case study was conducted to evaluate the prototype in a real-world environment. Results indicate improved factual grounding compared to an LLM-only baseline within the evaluated dataset, although the evaluation was limited to a small set of queries and a single institutional document collection.

Assessing organizational readiness for highperformance computing (HPC) adoption requires evaluation beyond hardware benchmarking, encompassing workforce capabilities, software maturity, data interoperability, and regulatory compliance. This paper presents a modular, rule-based decision-support framework implemented in Python that evaluates HPC maturity across five integrated dimensions and generates a phased migration roadmap through a weighted scoring model and recommendation engine. The framework employs a formally defined aggregation formula with configurable dimension weights, and its outputs are validated through a basic sensitivity analysis demonstrating score stability under weight variation. Demonstrated on two simulated organizational profiles-a mid-sized research institute and a public administration body-the framework identified critical gaps in workforce readiness and governance compliance, highlighting the role of non-technical factors in HPC transition planning and the practical value of transparent, reproducible maturity assessment for early-stage decision-making.

This paper presents TutorMe, an AI-assisted elearning chatbot platform designed to support students with lesson explanations, question answering, quiz generation, and learning-material navigation using curriculum-aligned, digitized resources. TutorMe is implemented as a modular web application integrated with a domain-restricted knowledge base built from structured PDF learning materials. The contribution is engineering-focused: we describe a reproducible design for a knowledge-base-grounded tutoring assistant, document key configuration choices (prompting strategy, retrieval behavior, and platform settings), and report a pilot offline evaluation using a rubric for manual assessment of correctness and groundedness. In an internal test set of 120 questions spanning biology, chemistry, physics, and mathematics at three difficulty levels, manual review showed that 80.0 % of answers were fully correct. Biology exhibited the lowest accuracy (60%) due to terminology imprecisions, while mathematics achieved 93.3%. We discuss limitations including hallucination risk, curriculum drift, privacy, and the need for teacher oversight, and outline steps toward deployment-grade validation.

Artificial Intelligence (AI) is becoming an important part of modern educational reforms, introducing innovative approaches and learning methods [2]. This study explores the application of artificial intelligence in the education system, examining whether a tool such as ChatGPT can generate pedagogically relevant and curriculum-aligned teaching materials. The research methodology is based on the analysis of the role of AI in education, focusing on the evaluation of the quality, accuracy, and pedagogical value of the content generated by ChatGPT-5. The study combines international research on the use of generative AI in schools with an analysis of materials created for teaching biology and mathematics in the sixth grade of primary school. The analysis included simple, detailed, and curriculum-aligned prompts to examine how different prompt types affect cognitive complexity, language clarity, and alignment with learning outcomes. The results show that all generated materials were factually accurate but differed in educational value. Tasks created using detailed and curriculum-aligned prompts demonstrated higher pedagogical relevance and contributed to deeper understanding and the development of critical thinking skills among students. The research confirms that thoughtful and responsible use of artificial intelligence can provide significant support to teachers in creating quality and educationally meaningful teaching materials.

With engineering architecture being shifted to meet the requirements of sustainable development, the need for optimized design solutions places precise engineering methods at the core of the contemporary industrial transition toward data-driven strategies. A timely conversion to lightweight components in drivetrain systems has led to the prominent use of high-strength polymer gears, establishing them as a critical point of interest in the field of power transmission. However, as the conversion to polymer gears relies on expensive and time-consuming laboratory testing, there is a standstill in evaluating the structural properties specific to polymer gear design. In addition, one of the major concerns in the development of polymer-based gear drives is linked with their operational performance and dynamic response under fault conditions influenced by surface wear. To address these difficulties, a framework for surface wear prediction is developed, enabling precise design optimization for specific drivetrain requirements. Computations of wear progression over multiple duty cycles are built upon the mathematical background of Archard’s wear theory, while internal changes in gear contact pressure distribution are constructed on Winkler’s surface model. The framework provides an innovative support for polymer gear systems, as it imports the three-dimensional (3D) scanning data of gear geometry, therefore enabling the analysis of actual flank surfaces with designated surface modifications and manufacturing errors. The framework’s effectiveness, confirmed by experimental validation, demonstrates a superior estimation of contact parameters and overall performance compared to traditional design methods, highlighting scalable solutions that contribute to ongoing industrial engineering objectives.

Purpose: This study examines the psychological implications of integrating artificial intelligence (AI) into judicial decision-making in criminal justice, including algorithmically supported risk assessment and sentencing decisions. It analyzes how AI-based decision-support systems influence perceptions of fairness, trust in judicial decisions, and decision confidence, as well as the emotional responses of judges, jurors, defendants, and victims. Methodology: The study employs a theory-driven and interdisciplinary conceptual framework grounded in psychological theories of decision-making, procedural justice, and affective processes. Through a critical integrative synthesis of legal, psychological, and ethical scholarship on algorithmic decision-making, predictive modeling, and risk assessment systems in criminal justice, the study examines their implications for human judgment, responsibility attribution, and judicial experience. Findings: The analysis demonstrates that AI-assisted decision-making can substantially shape psychological perceptions of justice and the legitimacy of judicial processes. Although algorithmic tools are often perceived as consistent and objective, their reliance on historical data may reproduce existing biases, thereby negatively affecting perceived fairness, trust in judicial outcomes, and decision confidence among legal professionals and trial participants. These findings indicate that the psychological impact of artificial intelligence extends beyond technical accuracy and plays a significant role in shaping perceptions of the legitimacy of judicial processes. Unique Contribution to Theory, Practice, and Policy: This study contributes to psychological theory by offering a systematic examination of the cognitive, affective, and evaluative processes associated with algorithmically supported judicial decision-making in criminal justice. In the context of judicial practice, the analysis demonstrates how uncritical reliance on AI systems may diminish judicial autonomy and obscure responsibility attribution in decision-making processes. From a public policy perspective, the findings contribute to the conceptualization of regulatory approaches oriented toward transparency, fairness, and trust in the use of AI in judicial decision-making.

This study explored the relationships between geological substrate and the structural and compositional attributes of mixed beech ( Fagus sylvatica L.), fir ( Abies alba Mill.), and spruce (Picea abies [L.] Karst.) forests on Mt. Konjuh in northeastern Bosnia and Herzegovina. Research was conducted on 81 experimental plots established across three dominant substrates: limestone, peridotite, and chert. Stand structure, diversity, and spatial organization were assessed using the Shannon diversity index, Pretzsch’s species profile index, Gini coefficient, and the Clark–Evans and Füldner indices. The analyses revealed consistent differences among substrates, suggesting that geological conditions influence forest structure and diversity. Higher diversity and vertical heterogeneity were generally associated with limestone, while stands on peridotite and chert exhibited simpler but more balanced structures. All forest types displayed a reverse J-shaped diameter distribution, indicating uneven-aged composition and ongoing natural regeneration. Spatial patterns showed a tendency toward clustering of beech and spruce and higher species mingling on limestone. Overall, mixed beech–fir–spruce forests on Mt. Konjuh appear to be stable ecosystems whose structure and diversity are shaped by an interplay of geological, edaphic, and ecological factors. The results highlight the relevance of site-specific and adaptive silvicultural approaches that account for local variability in substrate and stand conditions.

Norway spruce ( Picea abies (L.) Karsten) is one of the most economically important conifer species in Europe. Efficient utilisation and processing of its wood require detailed knowledge of its technical properties, as well as the most common wood defects that substantially affect both properties and utilisation. Given the crucial role of wood defects in the roundwood classification system, the primary objective of this study was to identify defects in Norway spruce and to analyse the influence of forest assortment characteristics (diameter and position along the stem) and tree attributes (diameter at breast height and position within the stand) on the size of wood defects. The research was conducted in Bosnia and Herzegovina, within a forest compartment of an uneven-aged, mixed beech and silver fir stand with spruce. Trees were felled and processed into assortments using a chainsaw, predominantly applying the cut-to-length method. After measuring the assortment dimensions, the occurrence of defects was assessed, and their sizes were determined. The analysis showed that, following knots, the most common wood defect was rot, followed by pith eccentricity, compression wood, scars, mechanical damage, and resin pockets. Statistically significant differences were found in the size of knots, ellipticity, and taper among different diameter classes of assortments (p<0.05), as well as assortment positions along the stem (p=0.0000). Also, a statistically significant difference was observed in the size of the knots and ellipticity in relation to both diameter at the breast height and tree position within the stand (p<0.05). Overall, the findings align with previous studies, confirming the higher quality of the lower stem section, as reflected in smaller defect sizes critical for roundwood quality classification, such as knots, rot, ellipticity, and taper.

The within-school gap in teaching has long been a primary focus of policymakers, researchers, and practitioners working to ensure equitable student outcomes. However, limited empirical research has examined the factors that can address this gap. This study examined the role of instructional school leadership in explaining variation in within-school creative pedagogies, controlling for various school-level contextual and teacher-related variables. The data come from 17 countries that participated in the 2022 PISA program. The analysis followed three steps. First, variables related to school and teachers were included in the regression. Then, instructional leadership was included in the analysis to examine its association with within-school variation in teachers’ use of creative pedagogies. The regression coefficients from each country were then combined in a meta-analysis to estimate the country-level effects. Across 17 countries, instructional leadership was generally associated with lower within-school variation in teaching quality, though this relationship was statistically significant in only three countries. These results point to a modest but potentially meaningful role for instructional leadership in reducing the gap among teachers in their creativity-oriented teaching practices and, in turn, promoting greater equity in student learning.

In terms of ichthyology, Bosnia and Herzegovina (BiH) is one of the most interesting parts of Southeast Europe, due to its rich biodiversity and high level of endemism. Despite its relevance, the entire territory has been poorly explored. Here, we provide an updated inventory of the current state of knowledge on fishes, including lampreys, from the freshwaters of BiH by hydrographic basin, with recent distributional data and updated taxonomic status reviewed and compared with previous lists. The checklist was compiled based on the existing scientific and grey literature, technical reports, scientific congresses, academic dissertations, and unpublished/personal observations. In total, 123 species including diadromous and euryhaline fishes have been documented in BiH freshwaters to date. Of these, 110 are primarily freshwater. In comparison to the last published monography (Sofradžija 2009), we present a 9% increase in species number (11 species), resulting mainly from taxonomic re-evaluations of existing taxa on the basis of new information and the adoption of a new changes in the taxonomic status of several species. Among the valid primarily freshwater species, 87 are native and 23 are non-native. A total of 38 endemic species have restricted distribution, and are threatened by numerous anthropogenic pressures. Four species are considered endemic only to BiH: Cobitis herzegoviniensis Buj & Šanda, 2014; Phoxinellus pseudalepidotus Bogutskaya & Zupančič, 2003; Telestes dabar Bogutskaya, Zupančič, Bogut & Naseka, 2012; and T. metohiensis (Steindachner, 1901). In total, 75 genera and 34 families are represented: Leuciscidae is represented by 37 species, the Salmonidae by 13, followed by the Cyprinidae, Cobitidae and Percidae, each with eight species. The native species richness follows a pattern similar to that observed in other southern European countries. A national list of endangered species has not yet been proposed to BiH and management strategies for their protection or conservation are also not implemented. Hopefully, this updated checklist will serve as a basis for future research aimed at understanding the origin and status of conservation of the BiH fishes diversity, and supporting effective management and conservation programmes.

Background/Objectives: COVID-19 severity is influenced by a complex interplay between host, viral, and environmental factors. Emerging evidence suggests that Neanderthal-derived genetic variants may influence the progression and severity of SARS-CoV-2 infection. This study aimed to evaluate the association between selected Neanderthal-derived variants and COVID-19 severity in the population of the Republic of Srpska, considering relevant clinical, sociodemographic, and lifestyle factors. Methods: This multicentric cross-sectional study included 402 participants, classified as healthy or SARS-CoV-2-positive individuals. A total of 378 COVID-19-positive participants were further stratified according to disease severity and hospitalization status. All individuals were genotyped for the Neanderthal-derived OAS3 rs1156361 (C/T) and LZTFL1 rs35044562 (A/G) variants. Detailed sociodemographic, clinical, and lifestyle data were also collected. Results: A higher frequency of the LZTFL1 rs35044562 AG genotype was observed among hospitalized patients compared with non-hospitalized individuals (36.8% vs. 20.9%; p = 0.005), while the AA genotype was more prevalent among non-hospitalized patients (77.3% vs. 63.2%, p = 0.015). Multivariable logistic analysis showed that carriers of the LZTFL1 AG genotype had a higher chance of hospitalization compared to AA carriers (adjusted OR = 1.372, 95% CI = 0.763-6.383, and p = 0.021). Hospitalized patients more frequently carried the combined CT (OAS3) and AG (LZTFL1) genotypes, supporting a potential synergistic effect. Several sociodemographic factors, including age, sex, education, employment, and urban residence, were also associated with COVID-19 severity, while no significant associations were observed in allele-based analyses. Conclusions:LZTFL1 gene polymorphisms may influence COVID-19 severity, with heterozygote-specific and combined risk effects observed. These preliminary findings are exploratory and require validation in larger cohorts, but may guide future studies and targeted interventions in high-risk groups.

The Open and Reproducible Musculoskeletal Imaging Research (ORMIR) community is a scientific community dedicated to promoting openness and reproducibility in musculoskeletal imaging, image processing, and computational modelling. In this perspective paper, we outline the motivations for conducting transparent research and provide practical guidelines to implement it. We start with defining open and reproducible research and describing the benefits and challenges of working transparently. Next, we redefine the outputs of a computational research study as—ideally—a combination of data, code, and a publication, recommend a folder and file structure that reflects these three study outcomes, and describe how to maintain and update such a structure during the study and at study publication. Finally, we emphasize that working in an open and reproducible manner is a learning process and the best way to acquire the necessary competencies is simply to start. Lay summary: The ORMIR community promotes openness and reproducibility in musculoskeletal imaging research. In this perspective paper, we explain why transparency matters and recommend how to conduct a computational study in an open and reproducible manner focusing on its three outputs: data, code, and publication. Finally, we highlight that the best way to learn these practices is simply to start.

Abstract The Open and Reproducible Musculoskeletal Imaging Research community is a scientific community dedicated to promoting openness and reproducibility in musculoskeletal imaging, image processing, and computational modeling. In this perspective paper, we outline the motivations for conducting transparent research and provide practical guidelines for implementing it. We start by defining open and reproducible research and describing the benefits and challenges of working transparently. Next, we redefine the outputs of a computational research study as—ideally—a combination of data, code, and a publication, recommend a folder and file structure that reflects these three study outcomes, and describe how to maintain and update such a structure during the study and at study publication. Finally, we emphasize that working in an open and reproducible manner is a learning process, and the best way to acquire the necessary competencies is simply to start.