Luxinnovation
Q: What do I get when I contact the one-stop-shop?
A: When you contact the AI Factory, you are assigned a single entry point: a Business Advisor who becomes your guide throughout the process. This person first clarifies your needs, checks eligibility, and helps you understand which services are relevant for your organisation. From there, you are guided through a structured journey that can include discovering AI opportunities, assessing your maturity, upskilling your teams, connecting with partners, securing funding, building and testing solutions, and eventually scaling them into production. The pathway is adapted to your company’s maturity level and sector. You gain coordinated access to expertise and infrastructure across the ecosystem.
Q: How do you help translate a business problem into an AI-ready use case?
A: The process typically begins with Fit 4 Digital AI, which serves as a structured diagnostic entry point. This assessment clarifies your business objectives, evaluates your readiness, and produces a practical roadmap aligned with your strategy. Once the problem is clearly framed, joint workshops are organised with AI experts from LNDS, LuxProvide, the University of Luxembourg, and LIST. These sessions refine the use case by identifying relevant data sources, suitable AI methods, technical feasibility constraints, and regulatory considerations. The outcome is not a vague concept note, but a concise and decision-ready use-case document. It describes the expected business value, required data, performance metrics, risks, compliance aspects, resource needs, and a concrete execution plan.
Q: Which funding instruments are realistically accessible for my company?
A: Funding options depend on your company size, project scope, and maturity level. Luxinnovation can support you throughout the application process to increase your chances of success.
Q: What kind of support do you provide during a funding application preparation?
A: Luxinnovation actively supports the structuring of your application. This includes verifying eligibility, matching your project to the most appropriate instrument, defining timelines, and clarifying formal requirements.
Q: What does an AI readiness assessment typically cover? And what do I get and the end?
A: An AI readiness assessment provides a structured evaluation of your organisation’s capacity to adopt AI effectively and responsibly. It examines strategic alignment, data and process maturity, governance and compliance frameworks, skills and operating model, infrastructure readiness, and risk management. Where relevant, it also considers regulatory frameworks. At the end of the assessment, you receive a maturity score, a prioritised portfolio of potential use cases, and a structured roadmap.
Q: What kinds of connections can you facilitate? Only in Luxembourg or in Europe as well?
A: The AI Factory facilitates connections at both national and European levels. Matchmaking goes beyond networking events. It covers experts, datasets, computing resources, compliance support, and co-innovation partners, with the objective of accelerating trusted AI adoption.
Q: Do you support only Luxembourg based companies, or also EU?
A: The AI Factory onboarding process and services are available to both national and European users. However, eligibility and prioritisation criteria may apply depending on the type of service and resource allocation.
Q: Is this only for startups?
A: No. The service catalogue is designed for startups, SMEs and large corporations. The journey is tailored depending on company size, sector, and AI maturity level.
Q: What are the typical timelines from first contact to project start?
A: Timelines vary depending on the complexity of the project, the maturity of the company, the availability and quality of data, and whether external funding or consortium partners are involved. In straightforward cases, where the problem is well defined, data is readily available, and no external funding is required, companies can move from initial contact to pilot launch within a few months. More complex projects, for example those requiring significant data preparation, regulatory analysis, multi-partner coordination, or EU funding applications, naturally take longer. The approach is designed to remain flexible: the pathway adapts to the ambition, risk level, and operational constraints of each project.
LNDS
Q: How can I identify datasets relevant to my AI use case?
A: Support applicants from shaping the initial idea to defining the data set required to build a valuable, effective model. At the same time, we assist data holders by assessing and exploring the availability of relevant data sets on the market, strengthening our role in data discovery. By engaging with data holders institutionally, we act as a trusted broker between data users and data owners, reducing the risks associated with data sharing.
Q: Can you help me access restricted or sensitive data, and under what conditions?
A: Access is possible, but always via secure environments, privacy safeguards, and controlled processes such as Secure Processing Environments (SPEs) (e.g., SPE built on top of Meluxina-AI), and our IPMS service that tackles data pseudonymization and identity matching to ensure data privacy — never through direct access to raw sensitive data. A specially tailored service for the preparation of data requests and support in the process is Data Access Request Review.
Q: What is the typical approval time for data access requests?
A: In case of request for re-use of public sector data, expect a formal response within 2–3 months under the Data Governance Act (DGA). For other cases, it depends on the rules set by the data holder.
Q: How do you assess data quality and suitability for AI training?
A: We help you determine whether your data is fit for AI training by combining quality checks and relevance assessments through our dedicated data quality and curation service.
Q: Can you support anonymisation, pseudonymisation, and data minimisation?
A: Yes, we can support anonymisation, pseudonymisation, and data minimisation through our IPMS service, which provides the controlled workflows and governance needed to apply these privacy‑enhancing techniques securely and in compliance with regulatory requirements.
Q: Can you help organisations understand which regulations apply to their AI use case?
A: Yes, we help organisations navigate the regulatory landscape by providing high‑level guidance on which frameworks may be relevant to their AI use case and by directing them to reliable resources for deeper exploration.
Q: How is data security ensured during experimentation?
A: Experimentation is performed only within secure, controlled environments and with privacy‑preserving transformations applied to the data where required. In addition, the data is encrypted at rest and in transit.
Q: Can you support data sharing across multiple partners?
A: Yes, we can support data sharing across multiple partners through our Identifier and Pseudonym Management Service (IPMS) service, which enables secure, controlled data exchange. The IPMS provides pseudonymisation in a decentralised manner, enabling data holders to pseudonymise and securely manage personal data at the source. It supports both individual and collaborative pseudonymisation projects, facilitating the secondary use of data in a secure environment for downstream analysis and related purposes.
Q: What documents do organisations need to prepare to demonstrate compliance with the EU AI Act?
A: The documentation required under the EU AI Act depends on the risk level of the AI system. We help organisations understand how these requirements may apply to their use case and, where relevant, provide access to templates or resources, when available, to support their preparation.
Q: Do you assist with ongoing data governance after deployment? [Do you offer post‑deployment data governance services?]
A: Introducing and maintaining the Data Governance mechanism is a permanent process that lives and changes following all the changes and functioning of the organization. Application of the AI model changes the view on data in organizations and requires greater focus, which is expected to be supported by adequate methods and tools. Accordingly, permanent work on improving the Data Governance process is expected even after the deployment of the AI model to ensure the application of remediation procedures and facilitate the regular review of the model. Data Management and Stewardship Support service helps organizations in achieving, improving and maintaining the Data Governance mechanism through consultations, trainings and co-development of policies, procedures, work instructions.
Q: Can you help organisations navigate regulatory requirements that apply to specific sectors?
A: Yes, we help organisations understand which sector‑specific considerations may influence their AI projects and direct them to the tools and references most relevant to their industry.
Q: Can you help organisations navigate regulatory requirements that differ between countries?
A: Our support covers the regulatory frameworks applicable in Luxembourg, including EU‑wide requirements. For country‑specific obligations outside Luxembourg, we recommend consulting the relevant national AI Factory or local specialists.
Q: Do you offer ongoing support as the AI system evolves?
A: Yes, we can continue to support organisations as their AI system develops, helping them stay aware of evolving requirements and pointing them toward suitable resources whenever new questions arise.
Q: Do you help clients prepare or transform their data for use in infrastructures such as MeluXina‑AI?
A: Yes, we support clients in preparing and transforming their data for use in infrastructures such as MeluXina‑AI through our data quality and curation service. This service ensure that datasets are properly cleaned, structured, validated, and optimised so they can be efficiently ingested and processed in high‑performance computing environments.
Q: Do you have processes to ensure privacy and security when operating in client or national infrastructure environments?
A: Yes, we have established processes to ensure privacy and security when operating in client or national infrastructure environments. This is supported through our Data Governance ELSI Expert Service, our IPMS service, and our DataOps team, which together provide a structured and compliant framework for secure operations. We apply rigorous privacy and security measures, including governed access controls, compliance‑aligned workflows, and continuous monitoring to ensure that all processing activities remain secure, traceable, and fully compliant with applicable regulatory and organisational requirements.
Q: How do you ensure your support activities do not introduce privacy or security risks?
A: We follow strict privacy‑first and security‑by‑design practices, including governed access controls, validated workflows, continuous oversight, and the application of appropriate privacy‑enhancing techniques. These measures ensure that every support activity is executed safely, transparently, and in full compliance with regulatory and organisational requirements.
Q: Do you have incident‑handling workflows for engagements involving client data (even if temporary)?
A: We have well‑defined incident‑handling workflows for engagements involving client data, including temporary access scenarios. These workflows are embedded within our ISO 27001‑certified Information Security Management System (ISMS) and our ISO 27701‑certified Privacy Information Management System (PIMS), ensuring that incidents are managed in a structured, secure, and fully compliant manner.
Q: We want to link several datasets from different sources - can you help?
A: Yes, our Identity Matching and Pseudonym Management Service is designed to facilitate privacy-preserving, record-linkable data sharing in Luxembourg, while Data Extraction, Enrichment and Merging service supports the secure provision and mapping of data from multiple providers while ensuring that data minimisation principles are respected.
Q: Who should I contact if I am facing issues with SPE access or staging data?
A: Any request related to access, installation of new software or packages, or other technical inquires must go through Service Desk.
Q: How much time does it take to publish a dataset in the L-AIF Datasets Catalogue?
A: Listing your datasets typically takes approximately 15-45 minutes per dataset, depending on the current documentation level, and you can rely on the support of our dedicated team that will guide you through the process.
Q: What are the benefits of publishing datasets in the L-AIF Datasets Catalogue?
A: Publishing datasets can help you increase strategic data visibility, improve data governance in your organisation, unlocke new revenue streams, and enable new collaborations.
Q: Does sharing metadata through L-AIF Datasets Catalogue expose my raw data?
A: No - as you onlu publish metadata and not the dataset itself, the underlying data remains fully under your control. You retain full ownership of your data as well as its access and usage rights, allowing you to make decisions on whether to grant access or now on a case-by-case basis.
LuxProvide
Q: When should I start using supercomputing resources in my AI project ?
A: You should begin using these resources as early as the design and prototyping phase. This allows you to generate rapid results and clearly define the next steps for your project.
Q: Which AI workloads are best suited for HPC environments?
A: Basically any AI workload is suitable. The primary limitation is the ability to train very large LLMs that would require a huge proportion of the machine’s total resources.
Q: Do I need in-house HPC expertise to use your services?
A: No. LuxProvide offers consultancy services to support you throughout your journey.
Q: Which AI frameworks and tools are pre-installed on MeluXina-AI?
A: The environment includes pre-installed tools for various needs: Inference: vLLM and ollama. Interface: OpenWebUI and Jupyterhub. Workflow: n8n. Storage: qdrant and milvus.
Q: How do you optimise compute performance and cost?
A: Optimization is managed through three main areas: Job Scheduling: We use SLURM for job scheduling where users request only the resources they need. Accurate resource requests reduce costs by avoiding unnecessary node allocations. Storage Tiers: Data movement is optimized using different tiers, such as "scratch" for temporary I/O heavy tasks versus permanent long-term storage.
Data Movement: Unlike hyperscalers, LuxProvide generally does not charge for downloading or uploading data, which is a significant cost differentiator.
Q: Can I combine your infrastructure with my own or cloud environments?
A: Interfacing Cloud and HPC is doable. This can be achieved directly or through VPN connectivity, among other more customized modes.
Q: How is access prioritised between users?
A: Access is managed via SLURM-managed scheduling queues based on job prioritization, resource availability, and fair distribution. Priority is defined by project allocations, with options for both dedicated nodes and on-demand resources.
Q: Do you support large-scale model training?
A: We support the training of medium to small-sized models as well as fine-tuning of existing models.
Q: What monitoring tools are available once my model is in production?
A: This is currently a work in progress
Q: Can you support continuous retraining and model drift detection?
A: This requires the integration of specific tooling, which is something we can work on together.
University of Luxembourg
Q: What AI training programmes are available for different maturity levels (awareness, practitioner, expert)?
A: We offer a full catalogue of AI training adapted to different maturity levels, from awareness sessions for beginners to practitioner and expert‑level programmes with hands‑on components.
Q: How do you help assess my organisation’s current AI skills gap?
A: The AI Light Maturity Assessment questionnaire offers an initial view of your organisation’s current capabilities. For a more in‑depth analysis, companies can apply for a Fit4AI or Skills Plang programme, which helps identify skills gaps and map them to appropriate training paths.
Q: Are training paths adapted for managers, business users, and technical profiles?
A: Yes, the programmes are tailored for different audiences, including managers, business users, and technical profiles.
Q: Do you offer modular or stackable learning paths over time?
A: Some training providers offer modular or stackable learning paths, while others provide standalone courses. Our catalogue is designed to give clear visibility across all available training options so that each company or individual can combine the programmes that best suit their needs and progressively build their own learning path over time.
Q: Are courses recognised with certificates or academic credentials?
A: Certification depends on the training provider. They may issue certificates of completion, and academic partners can award academic credits for courses taken as part of an accredited programme.
Q: Can training be customised for my sector or business context?
A: Customisation is possible through tailored recommendations and adapted learning paths, but the core training programmes remain the responsibility of the training providers. The AI Factory can share feedback with providers to reflect the needs and priorities expressed by industry.
Q: What delivery formats are available (online, hybrid, on-site)?
A: Courses are available online, on-site, or in hybrid formats, depending on the provider's offer.
Q: How do you ensure training content stays aligned with the latest AI standards and regulations?
A: Ensuring alignment with the latest AI standards and regulations is primarily the responsibility of each training provider. However, we curate the catalogue to maintain the highest level of quality, taking into account participant feedback and the evolving regulatory landscape. All AI Factory partners are committed to meeting rigorous quality standards, and we review the catalogue regularly to ensure it reflects current best practices and compliance requirements.
Q: How do you measure the impact of training on operational AI adoption?
A: Impact measurement is carried out through follow‑up with AI Factory clients, including a satisfaction questionnaire completed after the training.
Q: Do you support explainable and trustworthy AI?
A: Yes. We support the principles of explainable and trustworthy AI across our training activities. These themes are reflected in the programmes delivered by AI Factory partners, and we prioritise them when curating the broader training catalogue. While external providers are responsible for their own content, we include only courses that align with our overall focus on responsible and transparent AI.
LIST
Q: What is the LIST AI Sandbox?
A: It's an environment for testing the trustworthiness of AI systems and help organizations understand and mitigate risks.
Q: When should I use a sandbox instead of production testing?
A: A sandbox can be used at any stage of the product lifecycle, which other than being a good practice also addresses the continuous monitirng obligations under the AI act. However, in our research projects we typically use the sandbox before release in production
Q: How do digital twins support AI deployment? How does AI support digital twin development?
A: AI can support digital twin development in complementary ways. First, AI components can be directly embeded within the digital twin, for example thorugh predictive models that enhance simulation, forecasting and decision-making capabilities. Second, AI can be used to analyze data generated by the digital twin,
extracting insights, identifying patterns and improving performance over time. In addition, AI agents can assist in the design and configuration of digital twins, helping to generate tailored analytical and simulation services adapted to specific use cases
Q: Can you validate AI models under real-world, sector-specific constraints?
A: Yes, our sandbox is modular and adaptable by design to allow testing under any type of constraints. In particular it can be deployed on any infrastructure which allows companies for instance to conduct all testing locally. We are currently working with the university of Luxembourg on the Sandbox configurator whose purpose is to deploy sandboxes which are even more customized for the specific use case
Q: Can multiple AI models be integrated into one operational framework?
A: N.A.
Q: How do you ensure robustness and reliability of AI solutions?
A: We conduct testing across many different dimensions, spaning from bias testing to factual accuracy. It is important to highlight that all our tests take into consideration multi-lingualism: being luxembourg a highly multi-cultural and multi-linguistic country we want to ensure that the same quality of service is asured to all customers and citizens regardless of their background
Q: Can you support sustainability and green AI objectives?
A: Part of our research involves frugal AI which aims at optimizing the computation of large AI models, thus reducing energy consumption and environmental footprint
Q: Do you work with industrial and public-sector use cases?
A: Yes. We actively work with both industrial partners and public sector organizations. As a public research and technology organization, LIST engages in collaborative research, technology transfer and innovation projects that support industry development, strengthen technological capabilities and address societal challenges.
Q: How is testing documented and reported?
A: We are actively involved in the design of documentation and reporting which is user firendly across a diverse range of stakeholders including technical experts, risk anagers, legal experts, ethic experts, etc.
Q: Can results be reused for certification or compliance?
A: Yes. The purpose of our testing apart from helping companies mitigate risks, is to produce evidence which can be used for compliance and certification purposes. In particular we follow closely the evolution of AI reated regulations and standards in order to ensure that our documentation is in line with compliance requirements