Abstract
As ESG regulations gain relevance, namely following the entry into force of EU's Corporate Sustainability Reporting Directive, there is an increased requirement for value chain transparency. Organizations that are heavily based on data, Artificial Intelligence or Machine Learning, may significantly contribute to the ESG profiles of their upstream clients through the resources they spend on data storage and processing, model training or model serving. However, the lack of standardized mechanisms to report ESG-relevant indicators from these organizations hinders effective integration into their clients' ESG disclosures. This paper is motivated by an identified need to facilitate ESG indicator reporting by ML organizations in the value chain, and proposes an architecture to automate the monitoring and reporting process. This architecture is based on the principle of observability, and on the integration and effective use of open-source tools to monitor data processing pipelines and MLOps. By implementing an automated observability-based monitoring framework, ML organizations can provide actionable ESG data that align with regulatory requirements, while reducing reporting overhead for clients. We address how this architecture can be seamlessly integrated into existing operational workflows of ML providers and their clients, enhancing transparency, accountability, and compliance. The proposed architecture ensures accurate, real-time reporting and creates a scalable foundation for ESG aligned innovation in Machine Learning and adjacent domains. © 2025 IEEE.

Abstract
The increasing prevalence of ML in industrial environments is driven by the growing availability of userfriendly frameworks and industrial data. Manufacturing Execution Systems (MES) enabled easy data collection and utilization for decision support, namely for anomaly detection, quality control, or object detection/classification. However, models for new ML problems are often trained without regard for previous models or data, potentially wasting resources and hindering knowledge transfer. This is due to a lack of systematic methods for identifying and leveraging relevant prior knowledge. In this paper, we propose an approach designed to address this inefficiency by reusing previously trained models in new ML tasks. We reuse models based on data similarity metrics to create ensembles on-the-fly. This allows for accurate predictions on new data while minimizing the need for training from scratch. This approach has the potential to significantly reduce resource expenditure on data labeling and model training within industrial organizations.

Abstract
Explainable Artificial Intelligence (xAI) techniques are nowadays widely accepted as one of the paths towards addressing the interpretability and transparency issues of using black box models. Such techniques may allow to understand, to a certain extent, how or why a model produced a certain output, which may even help identify problems with the model or the data. As in many other domains, the use of xAI techniques in the context of manufacturing is seen as fundamental towards understanding model outputs, supporting informed decision-making, or enabling more human-centric approaches. In this paper, we specifically look at LIME, one of the most widely used approaches to xAI, and at how it needs to be adapted to the manufacturing context. Specifically, we show how the image permutations introduced by LIME might deceive the underlying model and generate poor explanations, and propose a methodology to address this issue. The specific use-case is on defect detection in the textile manufacturing industry.

Abstract
Ambient intelligence (AmI) refers to environments where smart devices, sensors, and AI-driven systems work seamlessly to enhance human interactions with their surroundings. Through the combination of real-time data, context-awareness, and adaptive learning, AmI enables environments to respond proactively to user needs, improving efficiency, comfort, and decision-making. However, since AmI systems are inherently human-centric and often operate autonomously, they must be designed with robust ethical, privacy, and safety considerations. Ensuring that these systems function reliably, fairly, and without harm is crucial, especially in sensitive domains like healthcare, security, and smart infrastructure. This work introduces a novel tool, conceptualized as an AmI Digital Twin, which allows developers to simulate or monitor AmI data streams, and develop and thoroughly test AmI applications before and during their real use. Built on a modular architecture leveraging technologies like React.js, Node.js, Kafka, Faust, MongoDB, InfluxDB, Grafana, and Docker, the platform ensures adaptability to different application environments, scalability, and ease of deployment. Besides the description of the tool itself, we provide some early validation results in common AmI tasks such as anomaly and concept drift detection. The tool is available in a public repository, and comes pre-packaged with a set of applications for AmI use-cases. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.