Abstract: With the deep integration of artificial intelligence (AI) into education, AI literacy has emerged as a core competency indispensable for pre-service teachers. However, the formation mechanisms and sustainable cultivation pathways remain underexplored. This study integrates the Technology Acceptance Model (TAM) and the Innovation Diffusion Theory (IDT) to construct a theoretical model where Individual Innovations (II) and Self-Efficacy (SE) serve as antecedents, Perceived Usefulness (PU) and Perceived Ease of Use (PEOU) as mediators, Behavior Intention (BI) as a proximal variable, and AI literacy (AIL) as the outcome variable. Through a questionnaire survey of 778 pre-service teachers, mixed empirical tests were conducted using Structural Equation Modeling (SEM) and fuzzy-set Qualitative Comparative Analysis (fsQCA). SEM results indicate that II and SE significantly and positively influence AIL through a chain mediation involving PE, PEOU, and BI. fsQCA further identifies four convergent high-AIL configurational pathways: "High-efficacy-practice-oriented" "High-adoption-intention-oriented" "High-innovative-qualities-oriented" and "Balanced-development-oriented". The study reveals that enhancing pre-service teachers' AIL involves diverse yet equivalent mechanisms, necessitating a shift beyond singular training paradigms. Based on these findings, the research proposes differentiated cultivation pathways, providing both theoretical foundations and practical references for teacher-training institutions to implement precise and sustainable AIL development.

Abstract: The immunoglobulins (IG) or antibodies and the T cell receptors (TR) are the antigen receptors of the adaptive immune responses (AIR) of the jawed vertebrates (Gnathostomata). IMGT®, the international ImMunoGeneTics information system®, was created in 1989 by Marie-Paule Lefranc (Laboratoire d’ImmunoGénétique Moléculaire (LIGM), Université de Montpellier and CNRS) to deal with and to manage the huge diversity of the IG or antibodies and TR. The founding of IMGT® marked the advent of immunoinformatics, a new science which emerged at the interface between immunogenetics and bioinformatics. For the first time, the IG and TR variable (V), diversity (D), joining (J) and constant (C) genes were officially recognized as ‘genes’ as well as were the conventional genes. The IMGT-ONTOLOGY CLASSIFICA-TION axiom and the concepts of classification have generated the IMGT nomenclature and the IMGT Scientific chart rules for assigning IMGT names to IG and TR genes and alleles of Homo sapiens and of an

Abstract: The optimization of herbicide application is one of the most important topics in Precision Agriculture, driven by both economic efficiency and ecological sustainability. Excessive herbicide use can lead to soil degradation, water contamination, and negative impacts on biodiversity, while also contributing to human health risks and climate-related concerns. Developing accurate, automated approaches for distinguishing crops from weeds is therefore essential to support sustainable agricultural practices. In this paper, a novel architecture for crops and weed segmentation in tobacco plantations is proposed: a U-Net variant which incorporates several specific design elements, including deep supervision, a Vegetation Global Context block, and a dual-headed output that separately predicts vegetation and crop masks. Weed regions are derived as the difference between vegetation and crop predictions, allowing the model to enforce logical consistency directly within a single framework, in contrast to other two-step approaches. The proposed architecture was evaluated using multiple modern encoder backbones. Experimental results demonstrate that this architecture not only improves segmentation accuracy compared to prior approaches, with best scores of 94.24% Dice for crop segmentation and 93.72% for weeds, but also significantly reduces inference time by avoiding multi-stage pipelines, making it much better suited for real-time deployment in field conditions.

Abstract: The approval of the selective NaV1.8 inhibitor Suzetrigine for acute pain has renewed optimism for developing novel analgesics, yet the clinical failure of its successor VX993 highlights the persistent difficulty of translating promising pain targets into effective therapies. This review examines why progress has been limited and how modern human centered approaches can reshape pain drug discovery. Human genetic studies from large biobanks demonstrate that genetically supported targets have a higher likelihood of clinical success. However, for pain, the relationship between genetic association and therapeutic efficacy is complex. Rare mutations in NaV1.7 and NaV1.8 strongly validate these channels as valid pain targets, yet common variant studies reveal little association with chronic pain risk, underscoring a polygenic and pathway level architecture rather than single gene causation. Human transcriptomic atlases of dorsal root ganglia (DRG) reveal extensive redundancy across NaV channel isoforms, helping explain the modest efficacy of selective NaV1.8 inhibition and pointing toward the need for multi target or pathway wide approaches. Multiomic analyses in osteoarthritis highlight additional pain generating mechanisms, including synovial inflammation, neuroimmune interactions, metabolic dysregulation, and osteoclast activity, along with the involvement of specific nociceptor subtypes. Human DRG electrophysiology and PK/PD modeling show that Suzetrigine achieves high NaV1.8 target engagement yet cannot fully silence nociceptors, and that central not solely peripheral NaV1.8 channel blockade may be required for robust analgesia. This helps explain the failures of peripherally restricted NaV1.7, NaV1.8 and TRPA1 channel blockers. Despite limitations, animal models remain essential for capturing integrated physiological responses and active drug metabolites not evident in vitro. Together, these findings support a more rigorous framework for target validation, integrating human genetics, multiomics, electrophysiology, and translational pharmacology to guide the development of next generation of pain therapeutics.

Abstract: Introduction: Heart failure (HF) is a commonly encountered fatal clinical condition. Catestatin is a peptide produced by the breakdown of chromogranin A and inhibits catecholamine secretion. The main goal of this study is to identify the difference between catestatin levels in patients at without and with different stages of HF. It is important to determine catestatin’s relationship with pro-B-type natriuretic peptide (proBNP) and left ventricular ejection fraction (LVEF), that are used to identify HF and as well as other laboratory findings in order to better understand the contribution of catestatin. Materials-Methods: Sixty HF patients with LVEF< %45 and 53 matching control patients of factors that can impact catestatin level were included in the study. Plasma samples of these patients and controls were simultaneously collected in tube containing a drop of aprotinin (proteinase inhibitor). Plasma catestatin levels were measured by using enzyme-linked immunosorbent assay method. Results: The study findings showed that catestatin levels of HF patients (45.46±16.69 ng/ml) were significantly higher than that of patients without (37.15±16.36 ng/ml) (t=2.69, p< 0.05) and that the catestatin level increases as the HF stage progresses (J-T=2.19; p< 0.05). Catestatin level is found to be correlated positively with proBNP (r=0.241; p< 0.05), and inversely with LVEF (r=-0.19; p< 0.05). The area under the ROC curve calculated in order to demonstrate catestatin’s diagnostic adequacy in heart failure was 0.635. Discussion: Catestatin is considered an indicator of HF and it seems reasonable to use it for diagnosis and follow-up as it increases with disease severity.

Abstract: Previous studies suggest that elevated competitive anxiety may increase the likeli-hood of injury. The present research aims to examine the role of competitive anxiety as a predictor of injury occurrence, frequency, and severity. A cross-sectional, correlational de-sign was conducted with 131 athletes, (mean age = 16.49 years), predominantly male. In-juries data were obtained through medical record review, and competitive anxiety was assessed using the Competitive Anxiety Inventory-2. Empirical frequency distributions, descriptive statistics, non-parametric tests, and logistic and ordinal regression models were employed. A high incidence of injuries was observed, although most were minor. Competitive anxiety was characterized by elevated levels of cognitive anxiety and self-confidence. Injured athletes exhibited greater overall competitive anxiety (r = .31, p < .001), with higher levels observed among those who sustained more injuries (ε² = .12, p = .001), and a very large effect was found in relation to injury severity (ε² = .17, p < .001). The occurrence of injury can only be predicted in 10.9–14.7% of cases through increased cogni-tive and somatic anxiety, whereas an increase across all dimensions of competitive anxi-ety predicts a greater number (13–14%) and severity (20.3–21.8%) of injuries. These find-ings underscore the importance of developing skills to manage competitive anxiety, par-ticularly its cognitive dimension and maintaining optimal levels of self-confidence in young athletes.

Abstract: This study aimed to investigate sex-specific behavioral differences in adult Wistar rats using a 3-minute Open Field Test (OFT), aligning with ethical guidelines emphasizing inclusive animal research for translational validity. While previous literature suggests female Wistar rats often display greater locomotor activity and central exploration, indicative of lower anxiety, these findings are not universal due to protocol variations. Fourteen Wistar rats (7 males, 7 females), aged 12 weeks, underwent the OFT in a controlled environment, adhering strictly to ethical protocols. Behavioral parameters assessed included locomotor activity, time in central zone, rearing frequency, grooming, and defecation. Data were analyzed using Student's t-test or Mann-Whitney test, with a significance level of p < 0.05. Results revealed no statistically significant differences between sexes for any analyzed variable (p > 0.05). Females exhibited numerically higher locomotor activity (50.43 ± 10.0 vs. 45.29 ± 12.3) and rearing frequency (17.43 ± 5.56 vs. 14.29 ± 4.15), whereas males showed numerically higher grooming (2.43 ± 2.51 vs. 0.14 ± 0.38) and defecation (1.71 ± 1.5 vs. 0.43 ± 1.13). These numerical trends, however, did not reach statistical significance. The findings align with the mixed results reported in the literature concerning sex differences in anxiety and exploratory behaviors within the OFT paradigm, particularly when using small sample sizes. Methodological limitations include the sample size and the absence of estrous cycle control, though the latter was reinterpreted for its ecological validity. Despite the lack of robust differences in spatial or general activity metrics, the observed trends in grooming and defecation hint at subtle sex-specific stress reactivity. This study contributes to methodological optimization by demonstrating the applicability of a brief 3-minute OFT protocol, which reduces animal exposure to experimental stress. Furthermore, it reinforces the critical importance of systematically including both sexes in behavioral research to ensure translational validity, even when statistically robust differences are not immediately apparent. Future research incorporating larger sample sizes and comprehensive hormonal monitoring is necessary for a more nuanced characterization of sex-specific behavioral responses in stress-related paradigms.

Abstract: Financial fraud is one of the biggest operational risks for financial institutions, generating significant financial losses and destabilizing the market. While machine learning models are good at predicting, their evaluation often relies on statistical performance metrics that don't directly translate into financial impact. This research develops an evaluation framework that integrates the costs of early fraud detection with predictive effectiveness and economic criteria for decision-making. Several supervised learning models (XGBoost, neural network, random forest, decision tree, and logistic regression) were trained and tested on an unbalanced dataset of credit card transactions. To measure the potential benefit of the models for financial institutions, the savings rate and expected loss were used, along with classic metrics such as F1 score, AUC-PR, AUC-ROC, recall, and accuracy. The economic results are highly sensitive to models with similar predictive capabilities. The ensemble methods, in particular, achieved the optimal balance between fraud detection capabilities and loss reduction, while models optimized solely for accuracy resulted in higher operating costs due to false positives or undetected fraud. The results indicate that the choice of fraud detection models should not be based solely on predictive accuracy, but also on cost asymmetry and risk tolerance. The proposed framework offers practical guidance to financial institutions seeking to align operational risk management and regulatory requirements with the implementation of machine learning, enabling risk-informed decision-making.

Abstract: The Santorini volcano, Greece, attracts global scientific interest and constitutes a top tourist destination. The 17th century BCE eruption, known as the Minoan event, was likely the largest ever occurred in the Holocene. The evaluation of an enriched collection of documentary sources combined with scientific observations showed that during historical times 14 small-to-moderate eruptive episodes were reported from the 2nd century BCE up to 1950 CE. Among them two little-known episodes occurring in 1667 CE and 1773 CE were uncovered and analyzed based on European documentary sources. For the first time a reliability score has been assigned to each one of the 14 episodes. The completeness of the recorded eruption history after the 14th century CE looks like ten times higher than in the previous period but it remains unclear whether this reflects real eruption rate or reporting incompleteness. The eruptions occurring after the 17th century CE are characterized by lower size, in terms of Volcanic Explosivity Index (VEI), than in the previous period. However, this may be due to the incomplete record of earlier eruptions of low VEI magnitude.

Abstract: Wireless communications face both opportunities and challenges due to the coexistence of 5G New Radio (NR) high-band, 5G mid-band, and 5G low-band technologies. Each technology uses both licensed and unlicensed spectrum to operate in separate frequency bands. For example, 5G NR uses the high-band of 24+ GHz, the mid-band of 2-6 GHz, or the low-band of less than 2 GHz, including the 5 GHz band via Licensed-Assisted Access (LAA). With the use of sophisticated coexistence mechanisms and optimization techniques, this 5G coexistence scenario in shared spectrum can be effectively managed. These strategies are essential for boosting network capacity, reducing latency, and ensuring fair spectrum use across different wireless technologies. This work provides a comprehensive system-level evaluation of multi-band coexistence and offloading strategies under realistic deployment assumptions. The simulation results confirm the effectiveness of the proposed model, showing that spectrum sharing and coexistence among these technologies deliver scalable and robust performance in heterogeneous service environments. This approach enables efficient load balancing across the entire network and highlights the need for additional features to achieve further performance gains.

Abstract: In this work, the so-called characteristic function method is proposed as a new approach to describe and interpret the diffusion process with interacting adsorbates in terms of the surface coverage. In this context, the intermediate scattering function is shown to be a characteristic function of probability theory which is also the generating function of the moments and cumulants of the jump probability distribution. The theoretical analysis carried out here consists of reviewing very briefly firstly the case of non-interacting adsorbates or very low surface coverages and extending secondly this method to low and intermediate surface coverages. As a direct consequence of this analysis, it is shown that the static structure factor is also a characteristic function of the adsorbate separation distances.

Abstract: The rapid advancement of artificial intelligence (AI) technologies presents unprecedented challenges for workforce management, particularly within human resource (HR) and people management functions that simultaneously face high AI exposure and serve as organizational architects of workforce adaptation. This article critically reviews and extends the emerging adaptive capacity framework introduced by Manning and Aguirre (2026), which measures occupation-level worker characteristics relevant for navigating job transitions following AI-induced displacement. While their framework advances understanding of differential workforce vulnerability, its occupation-level aggregation obscures critical within-function heterogeneity, particularly in HR domains where roles range from transactional administration to strategic business partnership. We extend the adaptive capacity framework by applying it specifically to HR functional areas, disaggregating people management occupations into distinct role clusters with varying exposure-capacity profiles. Drawing on strategic HRM theory, including the resource-based view and ability-motivation-opportunity frameworks, we develop a multi-level adaptive capacity model integrating individual, occupational, organizational, and institutional factors. Our analysis reveals that HR functions exhibit pronounced bifurcation: transactional and administrative HR roles demonstrate high AI exposure coupled with low adaptive capacity, while strategic HR business partners and organizational development specialists show moderate exposure with substantially higher adaptive capacity. Using paradox theory, we examine how HR practitioners must navigate the tension between facilitating organizational AI adaptation and experiencing their own occupational transformation. We also address equity implications, examining how differential adaptive capacity may interact with existing workforce inequities. The article offers both theoretical refinement and practical guidance for HR leaders, policymakers, and management scholars concerned with workforce resilience in an era of accelerating technological change.

Abstract: Background: Gastrointestinal graft-versus-host disease (GVHD) is one of the most severe complications of hematopoietic stem cell transplantation (HSCT), particularly in pediatric patients. It is frequently associated with treatment refractoriness and prolonged dependence on parenteral nutrition. Teduglutide, a GLP-2 analog, has shown promise in intestinal mucosal regeneration, but its use in GVHD remains limited. This case report describes the clinical experience with teduglutide in a pediatric patient with refractory GVHD-associated enterocolitis. Methods: We report the case of a 3-year-old child with primary immunodeficiency (Hyper-IgM Syndrome – CD40L deficiency) who underwent haploidentical HSCT and subsequently developed grade III gastrointestinal GVHD. The patient presented with severe chronic diarrhea, rectal prolapse, extensive intestinal inflammation, and nutritional failure despite standard immunosuppressive therapy. Teduglutide was initiated on a compassionate basis at a dose of 0.9 mg subcutaneously once daily. Results: Following initiation of teduglutide, the patient showed sustained improvement in stool consistency and frequency, reduced signs of intestinal inflammation, nutritional stabilization, and resolution of rectal prolapse episodes. Clinical response was maintained until the patient's death due to pulmonary infectious complications unrelated to gastrointestinal involvement. Conclusions: The positive intestinal response observed in this case supports the potential role of teduglutide as an adjunctive therapy for mucosal recovery in refractory GI-GVHD. Although encouraging, its use in this setting warrants further investigation through controlled studies, particularly in pediatric populations.

Abstract: Background: The presence of alkali species in ground-based spectra of comets is complex. The observed abundance ratios deviate from solar composition, suggesting alkali ejection from phenoxides reacting with carbon dioxide at the nucleus surface. Methods: Here we search for the emissions of the alkali in spectra of the coma and of the trail of Comet C/2023 A3 (Tsuchinshan-ATLAS) exploiting the double-fiber entrance of the very high resolution PEPSI spectrograph at the 8.4m Large Binocular Telescope. Results: Spectra sampling the nucleus yield $Na/K$ ratios about four times higher than the solar value, and even higher ratios sampling the trail. This fact excludes photodesorption as the main sodium source, leaving phenoxilation at the surface of the main nucleus and the trail mininuclei as the primary sodium source. Conclusions: The C/2023 A3 nucleus temperature and the faint KI line exclude potassium phenoxylation. For the first time, KI is detected in the trail of a Oort cloud comet, being potassium photodesorbed from the trail mininuclei. Sodium phenoxylation is at least six times more efficient than sodium photodesorption if the $Na/K$ ratio in the C/2023 A3 nuclei is chondritic. Trails composed of sub-km-sized mininuclei may be common features of Oort cloud comets.

Abstract: Burkholderia pseudomallei, the causative agent of melioidosis, presents significant challenges in both treatment and environmental decontamination. Bacteriophages, or phages, are increasingly being explored as potential diagnostic, therapeutic and biocontrol agents against this bacterial pathogen. Our recent investigation has shown that most B. pseudomallei genomes contained pro-phage(s) associated with specific tRNA gene loci, prompting us to explore these detectable pro-phages as sources of temperate phages, for further applications. Transcriptomic profiling of B. pseudomallei Bp82, a model strain that possesses three different prophages, revealed high expression levels of the integrase and certain transcriptional regulatory genes within its prophages during normal exponential growth. Using one of its temperate phages, namely φBP82.2, a P2-like phage, as a model, we investigated the lysogenic-lytic control mechanisms. Mutagenesis of the integrase gene, phiBP82.2_gp51, did not improve killing activity compared to the wildtype phage. In contrast, deletion of phiBP82.2_gp38, a putative transcriptional regulatory gene, and two-downstream hypothetical protein genes, phiBP82.2_gp36 and phiBP82.2_gp37, resulted in significant lytic improvement. We conclude that these genes play a crucial role in the lysogenic-lytic switch of φBP82.2, suggesting a new avenue for engineering temperate phages for future applications.

Abstract: Background: Antinuclear antibodies (ANAs) are frequently detected in patients with rheumatoid arthritis (RA); however, their prognostic relevance in predicting treatment escalation remains uncertain. Identifying biomarkers associated with earlier transition to advanced therapies may improve individualized disease management. Objectives: To evaluate the association of ANA status and titer levels with clinical characteristics, treatment trajectories, and time to biologic therapy initiation in patients with RA. Methods: In this retrospective cohort study, 223 patients with RA were stratified ac-cording to ANA status (112 ANA-positive, 111 ANA-negative). Baseline demographic data, disease activity (DAS28), and serological markers (RF, anti-CCP) were analyzed. Time to biologic therapy initiation, defined from the date of RA diagnosis to first bio-logic or targeted synthetic DMARD use, was assessed using Kaplan–Meier survival analysis and Cox proportional hazards regression. Multivariate models adjusted for relevant clinical covariates. Within the ANA-positive group, exploratory analyses compared low–moderate (1:80–1:320) and high (>1:320) ANA titers. Results: Baseline demographic and clinical characteristics were comparable between groups (all p > 0.05). ANA-positive patients more frequently initiated biologic therapy (48.2% vs. 24.3%, p < 0.001) and underwent multiple biologic switches (29.5% vs. 16.2%, p = 0.028). In multivariate analysis, ANA positivity was independently associated with earlier bi-ologic initiation (adjusted HR 2.14; 95% CI 1.32–3.46; p = 0.002), whereas RF and an-ti-CCP status were not significant predictors. In exploratory subgroup analysis, high ANA titers (>1:320) were associated with a lower hazard of biologic initiation com-pared with low–moderate titers (HR 0.24; 95% CI 0.06–0.98; p = 0.048). Conclusions: ANA positivity was independently associated with earlier initiation of biologic therapy in RA, supporting its potential incremental prognostic value beyond traditional sero-logical markers. The observed non-linear association between ANA titers and treat-ment escalation warrants cautious interpretation and validation in prospective studies.

Abstract: Background: Artificial intelligence (AI) is increasingly used to support tuberculosis (TB) screening and diagnosis, especially computer-aided detection (CAD) applied to chest radiography (CXR). The value of these programs depends not only on diagnostic accuracy but also on threshold calibration, integration into clinical workflow, and capacity for confirmatory testing. Methods: We conducted a narrative state-of-the-art review of AI applications relevant to TB screening and diagnosis. We synthesize evidence from World Health Organization policy documents, independent validation initiatives, and peer-reviewed studies re-porting diagnostic performance and real-world implementation outcomes. Results: CAD for CXR is the most mature AI application and is recommended by WHO for TB screening and triage among individuals aged ≥15 years in specific contexts. CAD-CXR can achieve sensitivity comparable to human readers, although performance varies by product, software version, population, and imaging conditions. Threshold selection is therefore a programmatic decision influencing referral volume and resource use. Inde-pendent benchmarking and local verification studies are essential to confirm performance and assess subgroup variability, including among people living with HIV and those with prior TB. Other AI approaches, including computed tomography (CT)-based imaging analysis, point-of-care ultrasound interpretation, cough or stethoscope sound analysis, clinical risk models, and genomic resistance prediction, are still at earlier stages and generally require further independent validation before routine programmatic use. Conclusions: AI has the potential to strengthen TB screening and diagnostic pathways, but impact should be evaluated using patient- and program-level outcomes rather than accuracy alone. Responsible scale-up requires local calibration, governance safeguards, and ongoing monitoring in real-world settings.

Abstract: This study extends previous research on key banking indicators during crisis periods by conducting a comprehensive comparative analysis of crisis and recovery periods for five major Canadian banks across five distinct economic crises from 1992 to 2020. It examines which market price indices significantly predicted daily closing prices during five crisis periods compared to their corresponding recovery periods using multiple linear regres-sion. Findings reveal that the Financials index shows consistently positive significance during crises and their recovery periods across all banks. During recovery periods, the significance of indicators changed toward sector-specific indices, interest rates, and market capitalization variables that were less important during crises. Banks exhibited heterogeneous recovery paths, with some normalizing quickly while others remaining sensitive to crisis-related predictors. These findings enhance understanding of banking sector resilience and inform portfolio management during recovery periods. The results indicate that recovery periods require distinct analytical approaches from crisis periods, with important implications for risk management and investment decisions.

Abstract: Background/Objectives: Wild boar (Sus scrofa) populations have expanded markedly across Europe, increasing human–wild boar encounters and hunting-related injuries. Existing knowledge derives mainly from isolated case reports or fatality analyses. This nationwide study characterizes injury patterns, management pathways, outcomes, and contextual risk factors associated with wild boar attacks during organized hunting in Germany and complements these data with a systematic literature review. Methods: Injured hunters were recruited via major German hunting journals. Structured physician-led telephone interviews captured demographics, hunting exposure, event characteristics, anatomical injury patterns, management strategies, complications, outcomes, and post-injury adaptations. Injuries were categorized as closed, open outpatient, or open inpatient. Descriptive statistics, Chi-square tests, linear regression models, and ANCOVA were applied. A systematic literature review (PRISMA 2020) supplemented the primary dataset. Results: A total of 101 hunters were included, predominantly experienced male dog handlers injured during close-range tracking operations. Lower-extremity penetrating injuries dominated. Significant associations with injury severity included situational context (p = 0.028), sex of the injuring boar (p = 0.023), and time to first help (p = 0.036). Open injuries requiring inpatient care frequently involved extensive soft-tissue destruction, vascular injury, surgical intervention, infectious complications, and prolonged recovery. Treatment duration strongly predicted work absenteeism across all injury categories, with progressively steeper regression slopes in more severe injuries (interaction p < 0.001). No fatalities occurred, and all participants resumed hunting. Most participants expressed need for improved first-aid training. Conclusions: Hunting-associated wild boar attacks constitute a distinct form of penetrating trauma with deceptively small external wounds and substantial underlying tissue damage. This study provides the largest structured clinical dataset to date and, combined with a systematic literature review, informs prevention strategies, first-aid preparedness, and surgical management in modern trauma systems.

Abstract: Industrial pouring processes operate under highly dynamic conditions where small deviations can lead to defects, scrap, and production losses. Although modern foundries are equipped with multiple sensors and visual inspection systems, most monitoring approaches remain fragmented, unimodal, and difficult to interpret. Furthermore, annotated anomalous samples in industrial settings are scarce, hindering the development of traditional methods. As a result, many critical pouring anomalies are detected too late or lack sufficient contextual information for effective decision making. In this work, we propose a multimodal framework for industrial scene characterization that unifies visual information and process signals through a Mixture of Experts (MoE) strategy. First, we deploy an ensemble of specialized modules that collaborate to identify regions of interest, assess pouring quality, and contextualize events within the production process, generating an interpretable description of pouring events. Second, we introduce a novel anomaly detection method for video multimodal data, combining a self-supervised transformer with an outlier-aware clustering algorithm. Our approach effectively identifies rare anomalies without requiring extensive manual labeling. The resulting information is structured into a digital-twin-ready representation, enabling seamless synchronization between the physical system and its virtual counterpart. This solution provides a scalable, deployable pathway to transform heterogeneous industrial data into actionable knowledge, supporting advanced monitoring, anomaly detection, and quality control in real foundry environments.