Apply

Next Generation Human-Centered Artificial Intelligence has the aim of designing AI systems that are not only accurate and efficient, but also understandable, trustworthy, inclusive and aligned with human and societal needs. The research of this PhD project will investigate advanced computer science methods for AI systems that can interact with people in transparent, adaptive and collaborative ways, supporting decision-making without replacing human responsibility. The candidate will study models, algorithms and evaluation methods for human-centered AI, including explainability, uncertainty awareness, fairness, robustness, personalization, user feedback, and human-in-the-loop mechanisms. Particular attention will be devoted to AI systems operating in complex socio-technical environments, where technical performance must be balanced with ethical, legal, cognitive and organizational requirements. The research may address application domains of high societal impact. The expected outcome is a new generation of AI methods and prototypes able to enhance human capabilities, promoting trust and accountability, and contribute to the responsible deployment of AI in society.

Making effective decisions in complex scenarios requires continuous coordination, information sharing, negotiating in situations of uncertainty and interpreting multimodal evidence. Although agentic AI is emerging as a promising technology for supporting such decision-making processes, current systems still fall short in terms of uncertainty awareness, reliable dynamics and trustworthy collaboration. Next-generation AI agents must exhibit advanced individual capabilities and function reliably as members of hybrid human–AI teams. In this context, they should be able to coordinate with others, communicate their own uncertainty, consider the uncertainties of their collaborators and decide when to act autonomously, refrain from acting or defer to humans or other agents. This PhD project focuses on reliability, uncertainty and the ability to defer in AI agents. The project aims to design, analyse and experimentally validate methods that enable agents to represent and quantify uncertainty in their predictions and reasoning processes; track how these uncertainties evolve over time and interaction dynamics. The overarching objective is to maintain reliable, calibrated behaviour in the face of distribution shifts, noisy or incomplete information, and heterogeneous interactions. Depending on the candidate’s background and interests, the PhD programme will combine theoretical work, with empirical studies in simulated and real-world scenarios, such as coordinated oncology patient management. This will contribute to the foundations and practice of safe and dependable agentic AI.

This PhD project aims to advance research on the YSocial (y-not.social) Digital Twin, a computational framework for creating dynamic digital representations of human populations that enable the generation of realistic synthetic societies and the in silico study of social phenomena. By integrating real-world data, agent-based modeling, machine learning, and network science, YSocial seeks to reproduce the evolution of social interactions, collective behaviors, and societal responses over time. Beyond the development of the platform itself, the project will exploit the synthetic data generated by YSocial to conduct virtual experiments on processes such as information diffusion, opinion formation, social polarization, and the impact of policy interventions. A key objective will be to evaluate the realism and predictive power of these simulations through systematic comparison with empirical observations.

This PhD scholarship supports methodological research in Explainable AI and Hybrid Decision-Making Systems towards safe and synergistic human-machine interaction and collaboration in the era of LLM based agents. Different lines of investigations are possible: i) Explainable AI (XAI):: Interpretability, post-hoc explainability, mechanistic explainability, and learning with uncertainty focusing also on XAI4Science ii) Trustworthy LLM: focusing on trustworthiness of collaborative and interactive learning in which human and LLM based agents engage in communication to integrate their distinct knowledge, mitigating hallucinations and sycophancy, RAG, and guidelines for high-risk AI systems iii) Human-AI Co-evolution: using the tools of complex systems science, to study the co-evolution between algorithms and society. In all cases, the investigation first will be grounded on some experimental setting and then generalized to generic design and deployment pipelines for safe Hybrid Decision Making Systems. Different experimental settings are available ranging from biomedical sector, financial, educational, mechanical engineering, and sentiment analysis and their choices will depend by the background and interests of the selected candidate. The research will be carried out mainly within the group of Data Science and Human Centered AI at SNS also in collaboration with other researchers of SNS depending by the domain of interest.

This PhD scholarships support methodological research at the intersection of Artificial Intelligence, Statistics, Computer Science, Engineering, and the Social Sciences. The research will be carried out within the Department of Excellence L’EMbeDS and will involve the three Institutes on which the Department is based: Economics, Management, and Dirpolis — Law, Politics and Development. The selected candidate will develop and apply advanced AI and data science methods driven by large-scale, complex, and heterogeneous data. The research may address theoretical, computational, and empirical challenges related to big data analysis, machine learning, statistical modelling, and trustworthy AI, with applications in Economics, Management, Law, and Political Sciences.

Agentic AI is an emerging paradigm that has the potential to reshape the way people work, interact, and organize activities, with implications that extend well beyond the workplace. As AI systems become increasingly autonomous and capable of pursuing complex goals, they raise important ethical, social, and governance challenges. This PhD project will investigate the ethical and social implications of the adoption of agentic AI, with the objective of developing strategies—including methods, tools, and governance frameworks—to ensure that human values, agency, and fundamental rights remain central in an AI-enabled society. The specific research focus will be defined in accordance with the successful candidate’s background, expertise, and research interests.

The PhD project will develop explainable neuro-symbolic agents able to support human deliberation while preserving control, responsibility, and trust. It will integrate neural language interpretation with symbolic reasoning, runtime verification, and cognitive mechanisms for inhibition, interruption handling, and post-error recalibration. Inspired by Stop/Ignore cognitive science paradigms, the research will define models and benchmarks for cognitive resilience in artificial agents. The framework will be validated in digital-twin scenarios for local energy communities and human-centred decision support.

This PhD scholarship supports research on novel algorithms, methodologies, and theoretical foundations for Artificial Intelligence, with a particular focus on Deep Learning. Special emphasis will be placed on AI technologies with significant societal impact, including privacy-preserving machine learning, personal data protection, trustworthy and explainable AI, fairness and bias mitigation, security and robustness against adversarial threats. The scholarship aims to foster the development of next-generation AI systems that are not only effective and innovative, but also reliable, secure, and aligned with fundamental societal values.

Modern AI systems face increasingly pressing challenges: their environmental footprint, their difficulty adapting to new data without forgetting old knowledge, and their reliance on centralized infrastructures that raise concerns around privacy and resilience. Green AI is concerned with reducing the computational and energy cost of machine learning, from training large models to deploying them efficiently on resource-constrained devices. This topic includes questions such as how to design models that achieve strong performance with fewer parameters, how to lower the energy consumed during training and inference, to mention a few. Continual and decentralized learning studies how models can adapt to new data over time without forgetting what they have previously learned, in settings where this learning happens across networks of devices without a central coordinator. Most current systems are trained once on a fixed dataset and rely on centralized architectures; making them work in non-stationary, open-ended, and distributed environments raises questions around privacy, communication efficiency, robustness, and knowledge retention, especially when the network itself is dynamic, with devices and connections appearing and disappearing over time. These two directions naturally interact but can also be pursued independently. The research focus will be defined together with the selected candidate.

Traditional machine learning focuses on correlations rather than causal relationships. Shifting toward causality is one of the most promising directions for AI that is more robust, interpretable, and practically useful. This research investigates causal learning frameworks that go beyond observational data, leveraging decentralized interventions to actively discover and refine causal knowledge in real-world settings. Pervasive computing environments, spanning smartphones, wearables, and IoT sensors, represent a natural substrate for this investigation. The heterogeneity and scale of device-generated data, combined with the ability to perform interventions across a distributed network of devices, create both unique challenges and opportunities for causal reasoning, making them an ideal case study for grounding causal AI in practice. In these settings, causal knowledge can improve decision-making and adaptive behavior, for instance by enabling systems to generalize across devices and environments, act robustly under uncertainty, or understand the consequences of their actions rather than merely reacting to observed patterns. Depending on the candidate’s background and interests, research activities may include theoretical modeling of causal discovery and inference in distributed settings, algorithm and system design for active causal learning on pervasive devices, using causal representations to improve planning and generalization, and experimental evaluation through simulations and real-world deployments.

3D digital assets are becoming increasingly common. Online platforms and accessible modeling tools allow artists, designers, and independent creators to produce, showcase, and share 3D models easily. Recent advances in generative AI, including text-to-3D and image-to-3D generation, further accelerate this trend by enabling 3D assets to be produced from simple prompts. Also buildings and structures benefits of these new technologies, both from a design and an optimization point of view. Assessing aesthetics of 3D models is a new and challenging task, because aesthetic traits goes behind visual appearance of an object involving aspects like intended use, presentation, composition, emotions, and other ones. In this PhD project we want to focus on the new task of aesthetic assessment of 3D objects developing novel deep learning architecture and solutions to modeling aesthetic traits, evaluate them, and to provide feedback and suggestions to artists and designers to support their design goals and improve their expressiveness.

This PhD project investigates the role of Large Multimodal Models in enabling Symbiotic AI for Earth Observation. The main goal is to understand how the most promising language and vision-language models can support users in interpreting, querying, and exploiting geospatial data more effectively. Earth Observation produces massive volumes of heterogeneous information, including satellite imagery, maps, time series, metadata, and textual reports. However, extracting meaningful knowledge from these sources still requires advanced technical expertise. This research aims to bridge that gap by studying how multimodal AI systems can understand geospatial intent and collaborate with humans in complex analytical workflows. A central objective is the design of a GeoAgent: an intelligent assistant capable of interpreting user goals, reasoning over spatial and temporal context, selecting appropriate Earth Observation data and tools, and producing reliable, explainable outputs. The project will analyze current Large Language Models and multimodal architectures, evaluate their limitations in geospatial reasoning, and explore methods to improve grounding, accuracy, and trustworthiness. By combining AI, remote sensing, and human-centered interaction, the research will contribute to more accessible, transparent, and effective Earth Observation systems, supporting applications such as environmental monitoring, disaster response, agriculture, and climate analysis

This PhD project explores methods and governance frameworks to ensure fairness, transparency, and regulatory compliance of AI algorithms embedded in socio-technical systems, with a specific focus on Civic Digital Twins. It aims to develop metrics, auditing approaches, and design guidelines to detect and mitigate bias, enhance accountability, and support trustworthy decision-making in public-sector, citizen-centric environments.

This PhD project investigates methods to improve the robustness and compliance of multi-modal AI systems for environmental pollution sciences. It focuses on robustness to distribution shifts, degraded inputs, and out-of-distribution conditions, while also addressing uncertainty quantification, explainability, fairness, and support for human oversight in heterogeneous environmental settings.

This topic concerns the design and deployment of advanced machine learning methods across cloud and edge environments to achieve scalable computation, low-latency inference, and efficient use of distributed resources. It emphasizes the trade-off between centralized model training and coordination in the cloud and localized execution at the edge for real-time, privacy-aware, and bandwidth-efficient intelligent distributed systems

The PhD research will address the design and training of next-generation multimodal foundation models, with particular focus on robust generalisation across diverse real-world distributions. The candidate will investigate architectural innovations in vision-language transformers (including cross-modal attention, connector design, and inference-time scaling) and study how training data composition, instruction tuning, and self-reflective reasoning affect out-of-distribution performance. A central thread will be the analysis and mitigation of fairness-related failure modes, framing trustworthiness as a first-class design constraint rather than a post-hoc correction. The research will have applications spanning visual question answering, retrieval-augmented generation, embodied perception, and multimodal agents, contributing to the broader agenda of safe and reliable foundation models.

The PhD research will tackle the problem of editing deep learning models using task arithmetic, with particular focus on the linear regime. The candidate will study task composition, updating, learning, and unlearning in deep learning architectures, primarily attentive models and large-scale transformers. Additionally, multi-objective task fusion exploiting Pareto frontiers will be investigated, with applications to both LLMs and vision models.

The growing adoption of generative artificial intelligence models (LLMs) is profoundly transforming the production, translation, and dissemination of specialised texts. However, these systems generate content based on statistical patterns and are not designed to ensure terminological consistency and conceptual precision, which are fundamental elements in scientific and technical communication. This research project aims to investigate the relationship between terminology, knowledge representation, and generative language models, to understand how LLMs manage specialised terms, conceptual relations, and disciplinary definitions. In particular, the study will examine whether and how the integration of structured terminological resources (termbases, ontologies, and domain-specific glossaries) can improve the semantic quality and terminological consistency of texts generated or translated by AI systems. Adopting an interdisciplinary approach that combines terminology studies, computational linguistics, translation studies, and data science, the project will analyse corpora of specialised texts generated by LLMs, conduct experiments with terminology-guided generation, and develop methodological models for terminology-aware AI systems.

This PhD scholarship focuses on pervasive, trustworthy, and sustainable artificial intelligence, combining advanced algorithms with security, privacy, energy efficiency, and social value. The project has two main directions. The first applies AI to precision medicine and pharmacological research, using NLP, Multi-Agent Systems, optimization methods, and Knowledge Graphs to integrate biomedical knowledge, select patient cohorts, predict Drug-Target Interactions, and support clinical decision-making through interpretable and trustworthy models. The second investigates AI-native architectures for 6G networks and the Cloud-Edge Continuum, enabling the distributed execution of Large Language Models and World Foundation Models in complex cyber-physical and IoT environments. The research will address cognitive scheduling, context-aware orchestration, low latency, energy sustainability, and Privacy-Preserving AI. Overall, the project connects the generation of scientific knowledge with its secure and efficient deployment, producing socioeconomic benefits through more resilient services, lower operational costs, and improved quality of support systems.

The integration of multi-omics data is currently one of the most complex computational challenges for decoding the nonlinear interactions that transcend the central dogma of biology. The heterogeneous nature and extremely high dimensionality of genomic, transcriptomic, and epigenomic profiles make traditional statistical approaches insufficient, requiring the development of next-generation Artificial Intelligence frameworks. The objective of the project is the design of Machine Learning architectures and the development of new algorithms specifically engineered for biomarker discovery. From an algorithmic perspective, the research will focus on implementing methods capable of increasing the analytical sensitivity and validity of biomarkers in a biological context characterized by a high level of background noise, intrinsic to the heterogeneity of the analyzed sources, namely malignant neoplasms. The computational tools developed will need to ensure high generalizability, operating across different biological matrices and resolution scales: from the analysis of solid tissues to complex single-cell datasets. Finally, the developed frameworks will be validated on real tumor cohorts, with the translational objective of improving early diagnosis and ensuring effective patient stratification.

Current mHealth systems generate vast amounts of multimodal data, yet their effective integration into reliable, personalized, and clinically meaningful decision-support tools remains largely unresolved. This PhD project aims to develop advanced Artificial Intelligence (AI) methods for extracting actionable insights from heterogeneous longitudinal data, including physiological signals, physical activity, sleep, environmental exposures, behavioral information, and patient-reported outcomes collected through mHealth technologies. The research will address key challenges in next-generation mHealth systems, including multimodal data fusion, missing and noisy measurements, longitudinal and real-world data modeling, personalization, explainability, trustworthiness, uncertainty quantification, and privacy-preserving learning. Novel approaches based on foundation models, causal AI, digital biomarkers, and adaptive learning will be investigated to support early risk detection, health trajectory prediction, and personalized interventions. Applications will span healthcare and healthy aging, sports performance and human monitoring in extreme environments and will be based on a multi-disciplinary approach with direct interaction with clinicians. The expected outcomes include novel AI methodologies, validated digital biomarkers, and intelligent mHealth solutions capable of supporting data-driven decision-making and personalized health management across diverse real-world scenarios.

This PhD scholarship aims to support advanced research at the intersection of Artificial Intelligence, education, and learning sciences. The selected candidate will focus on the design, development, and evaluation of AI-based tools for learning environments, with particular attention to their pedagogical effectiveness, usability, and functional impact in real educational contexts (e.g. different academic levels and professional training). The research will explore how AI technologies, such as adaptive learning systems, intelligent tutoring systems, generative AI, and learning analytics, can enhance learning outcomes, personalize educational experiences, and support teachers and institutions in data-informed decision-making. The candidate will also investigate the broader pedagogical implications of AI adoption, including changes in teaching methodologies, student engagement, assessment practices, and ethical considerations. The project combines technical development with empirical educational research, requiring interdisciplinary collaboration across computer science, cognitive science, and pedagogy. The researcher will include the exploration of innovative AI-driven educational solutions and to the scientific understanding of how these technologies reshape learning processes.

This PhD project will investigate AI-based approaches for reliable emotion recognition using non-invasive biometric sensors. Timely recognizing emotions is crucial to support the productivity and well-being of cognitive workers. To this aim, recent research has proposed approaches for sensor-based emotion detection during cognitive tasks, with promising results. However, limitations exist due to individual physiological differences, which we aim to address in the current PhD research. Specifically, we will investigate machine learning and deep learning architectures for training emotion recognition models specific to groups of people with similar physiological profiles. We will evaluate approaches on data collected from different application scenarios, with particular focus on software developers involved in programming tasks. The research will include a comprehensive evaluation of the proposed approaches as well as an investigation of the correlation of emotions and physiological profiles with cognitive aspects, well-being, and productivity of knowledge workers.

Caregiving for vulnerable people — older adults, chronically ill patients, or individuals with disabilities — exposes caregivers to a significant psychological burden, often accompanied by social isolation and difficulties in accessing services. In this context, chatbots based on large language models represent an emerging opportunity to provide a first level of listening, guidance, and emotional validation. The research aims to investigate how the pragmatic features of dialogue — tone, communicative style, turn-taking management, and the ability to adapt to the user’s emotional context — influence the perception of social presence, trust, and relational alliance. The risks related to emotional dependence on the system, the inappropriate replacement of human support, and the ethical implications of empathic simulation will also be examined.

The project explores the socio-cognitive dynamics of human–AI interaction, with particular attention to the processes through which individuals interpret, evaluate, and regulate their relationship with artificial agents. In particular, it analyses how the attribution of intentionality, competence, reliability, and relational meaning to artificial agents is intertwined with processes of human social cognition. By integrating psychological and behavioral data with psychophysical measures, the project adopts a multilevel perspective aimed both at gaining an in-depth understanding of the dynamics involved in relationship construction and at informing AI systems capable of adapting their responses to such dynamics. This perspective is particularly relevant in educational and socio healthcare contexts, where the quality of interaction, the multilevel analysis of human behavior, and the responsiveness of the artificial agent can significantly affect relational, decision-making, and support outcomes.

Digital health and rehabilitation are increasingly shaped by the availability of heterogeneous data sources, including clinical records, imaging, biomechanical signals, wearable sensors, patient-reported outcomes, and telerehabilitation platforms. The overall goal of this PhD project is to develop reliable AI methods for integrating and analyzing multimodal health data to support personalized assessment, prognosis, and rehabilitation planning. The research will focus on predictive models capable of capturing complex relationships among motor, cognitive, physiological, and contextual variables, with the goal of improving clinical decision-making and patient outcomes. Particular attention will be devoted to trustworthiness, including explainability, robustness, fairness, privacy preservation, uncertainty estimation, and compliance with ethical and regulatory requirements. The project will investigate machine learning and deep learning architectures for multimodal fusion, temporal modeling, and adaptive prediction, addressing challenges such as missing data, limited sample sizes, class imbalance, and variability across patients and clinical settings. Expected outcomes include interpretable and clinically meaningful models for rehabilitation monitoring, outcome prediction, and treatment personalization. By combining methodological innovation with real-world digital health needs, the project seeks to advance AI-based rehabilitation services that are transparent, safe, and usable by clinicians and patients.

Cardiovascular diseases remain among the leading causes of morbidity and mortality worldwide, demanding earlier detection, more accurate risk stratification, and personalized management strategies. This PhD project aims to develop trustworthy artificial intelligence methods capable of integrating heterogeneous cardiovascular data, including clinical records, imaging, electrocardiographic signals, wearable sensor streams, laboratory biomarkers, and lifestyle information. The project will investigate multimodal learning architectures that combine temporal, visual, and structured data to identify disease patterns, predict clinical outcomes, and support decision-making across prevention, diagnosis, and follow-up. Particular attention will be devoted to model interpretability, robustness, fairness, and clinical usability, ensuring that predictions are transparent and aligned with medical reasoning. By combining advanced machine learning, deep learning, and explainable AI techniques, the research will address challenges such as missing data, data imbalance, patient heterogeneity, and domain shifts across healthcare settings. The expected outcome is a set of intelligent, generalizable, and clinically meaningful models for cardiovascular health assessment. The research will explore validation strategies for real-world deployment in diverse clinical environments and patient populations. Ultimately, the project seeks to contribute to precision cardiology by enabling data-driven tools that support clinicians, empower patients, and improve cardiovascular care pathways

The PhD project aims to develop advanced vascular and cardiovascular modelling approaches based on physiological signals, including but not limited to ultrasound imaging acquired through non-invasive techniques. The research will investigate how multimodal data, including ultrasound-derived biomarkers, arterial pulse waveforms, electrocardiography, and blood pressure measurements, can be integrated to generate patient-specific models of the cardiovascular system. A core aspect will be the application of AI to the analysis of both ultrasound images and physiological signals, as well as to multimodal data integration. Deep Learning and advanced image and signal processing techniques will be employed to analyse ultrasound image sequences, blood pressure waveforms, and electrocardiographic signals, enabling the detection and quantification of early and advanced alterations in cardiovascular morphology, tissue composition, wall motion, flow dynamics, and atherosclerotic plaque burden. AI-based multimodal frameworks will combine imaging-derived features with clinical data and physiological signals, enabling a more comprehensive characterization of cardiovascular structure and function. The project will further explore data-driven and physics-informed modelling strategies to improve model personalization, identify novel cardiovascular biomarkers, and support risk assessment. The expected outcome is the development of innovative AI-enabled tools for quantitative cardiovascular assessment, contributing to precision medicine through improved diagnosis, risk stratification, and longitudinal patient monitoring.

The work funded by these scholarships will be developed within the Istituto Nazionale di Fisica Nucleare (INFN). INFN was funded about 70 years ago as the national institution devoted to research in fundamental physics, oriented to nuclear, particle and high energy physics, the disciplines describing the elementary building blocks of Nature and their interactions. Nowadays, INFN research extended towards wider directions including, for instance, multi-messenger astrophysics, cosmology, particle physics at colliders, neutrino physics, formal development of string theory and applications to society, including the fields of medicine and cultural heritage. In all such directions, the workflow generally includes the development of an underlying theory, the simulation of synthetic data (Monte Carlo simulations), the design and, often, the construction of experiments, the collection of real data and the processing and analysis of such data. Moreover, in several fields, such as particle physics and astrophysics, datasets became so large (and often sparse) to drive the entire field in a so-called big-data phase. Modern Machine Learning (ML) techniques can help the optimization, the efficiency and the speed up of each of the steps of the aforementioned workflow, especially in this big-data phase. This opens up new and promising research directions, all relevant for INFN research. Moreover, INFN can serve as a rich data laboratory to challenge new ideas and algorithms in the many directions of ML, from the explainability of ML models to low-latency algorithms to be used for real data analysis.

This PhD project aims to develop innovative data-driven and physics-informed computational methods for the prediction and analysis of urban microclimate phenomena. Urban environments are characterized by strongly multiscale and multiphysics processes involving fluid dynamics, heat transfer, radiation, and pollutant dispersion. High-fidelity numerical simulations can provide accurate predictions but are often computationally prohibitive for real-time applications, uncertainty quantification, and large-scale scenario analysis. The research will investigate novel SciML approaches that combine physical laws, numerical simulations, and heterogeneous observational data to construct efficient, reliable, and interpretable surrogate models. The PhD candidate will focus on the development of reduced-order modeling techniques, physics-informed neural networks, operator-learning frameworks, and hybrid data-assimilation strategies for computational physics applications. Particular attention will be devoted to integrating domain knowledge into machine learning architectures, enhancing model generalization, quantifying predictive uncertainty, and enabling robust predictions under sparse and noisy data conditions. By bridging computational physics, machine learning, and data assimilation, the PhD research will advance the state of the art in scientific machine learning. This PhD project will be supported by the prestigious ERC project DANTE – Data Aware Efficient Models of the urbaN microclimate (GA 101115741)

The research project is grounded in the growing gap between the rapid development and deployment of Artificial Intelligence agents (“AI Agents”)—algorithms capable of autonomously negotiating, making operational decisions, and concluding agreements—and chatbots, understood as computational tools designed to simulate conversational interaction with humans, on the one hand, and the still limited level of awareness and understanding of these technologies among scholars, policymakers, public and private institutions, and citizens at large, on the other. Against this background, the project examines digital assistants and conversational AI from a legal perspective, with particular attention to consumer-facing chatbots and AI agents. It investigates key questions concerning the validity of contracts concluded through AI systems, as well as the legal requirements of transparency, fairness, informed consent, and the effective exercise of consumer rights. The project also addresses the risks these systems may pose to end-users, including emotional dependence, behavioural manipulation, and forms of digital addiction, especially where such technologies are deployed at scale. By analysing these issues within contemporary regulatory frameworks, the project aims to clarify how existing law can respond to emerging forms of human-AI interaction and to identify areas where further legal safeguards may be needed.

The PhD project will develop methodologies and governance instruments for the effective implementation of the European AI Act across public and corporate contexts. It will integrate legal analysis, organizational design, and policy innovation to address challenges related to AI procurement, risk governance, compliance management, and accountability. Through case studies and co-design activities with stakeholders, the research will define operational frameworks that bridge regulatory requirements and organizational practice. The resulting models and tools will be validated in real-world adoption scenarios to support trustworthy, transparent, and sustainable AI deployment.