Designing the Future of AI-Enabled Research: Community-Driven Insights for the NAIRR Portal

Designing the Future of AI-Enabled Research: Community-Driven Insights for the NAIRR Portal Sandra Gesing San Diego Supercomputer Center University of California San Diego La Jolla, CA, USA [email protected] Maytal Dahan Texas Advanced Computing Center University of Texas, Austin Austin, Texas, USA [email protected] Linda Hayden Elizabeth City State University Elizabeth City, North Carolina, USA LBHA[email protected] Claire Stirm San Diego Supercomputer Center University of California, San Diego La Jolla, California, USA [email protected] Janae Baker San Diego Supercomputer Center University of California, San Diego La Jolla, California, USA [email protected] Abstract—The National AI Research Resource (NAIRR) Pilot is a U.S. government-led initiative designed to broaden access to the computational, data, and educational resources needed for artificial intelligence research and training. Coordinated by the National Science Foundation, in partnership with over a dozen federal agencies, the Pilot serves as a testbed to inform the design of a future full-scale NAIRR by exploring models for access, usability, and public-private collaboration. Between Fall 2024 and Spring 2025, SGX3, the Center of Excellence for Science Gateways conducted a mixed-methods study about envisioning the future NAIRR Portal involving a national survey, eight structured focus groups and a two-day design thinking workshop, collectively engaging more than 1,200 stakeholders from different backgrounds. The findings highlight enduring challenges such as fragmented access to computational resources, limited training opportunities, and steep entry barriers to adopting AI/ML tools. The participants articulated a strong shared vision for the NAIRR Portal: a user-centered platform offering personalized access, intelligent AI assistants, integrated FAIR data practices, multilingual support and embedded frameworks for responsible AI. This vision also emphasized the importance of fostering collaboration and community through accessible features. We present concrete technical and community-oriented design recommendations, including support for real-time metadata auditing, federated authentication, and modular architecture to enable adaptable, secure, and intelligent research workflows. Taken together, these insights offer a community-validated blueprint for building a scalable, equitable, and trustworthy national AI portal that can support the next generation of researchers and innovators. Index Terms—Science Gateways, NAIRR portal, NAIRR, high performance computing, machine learning, artificial intelligence. INTRODUCTION The National Artificial Intelligence Research Resource (NAIRR) Pilot [1] is a proof-of-concept initiative funded by the National Science Foundation, in collaboration with twelve other federal agencies and over two dozen nongovernmental partners. It aims to support inclusive and innovative AIenabled research and education by integrating computing, data, and training resources. Central to this vision is the development of a user-centered, intelligent portal that enables researchers, educators, students, and institutions, regardless of size or resource availability, to access AI tools, share knowledge, and collaborate on research problems in a secure and scalable environment. To investigate how such a portal could best serve its user base, a research project was carried out between Fall 2024 and Spring 2025, led by SGX3, the Center of Excellence for Science Gateways [2]. The project used a comprehensive mixed methods approach to gather perspectives and ideas from the wider research and education communities. This included a national survey, eight structured focus groups, and a twoday design thinking workshop. These activities collectively engaged more than 1,200 individuals and generated qualitative and quantitative data that shaped the understanding of the requirements of the portal design [3]. The purpose of this research was not only to inform technical feature development, but also to explore how a federated, community-driven platform can support responsible AI adoption. Participants from a variety of institutions, including doctoral universities, research centers, and small colleges, identified critical challenges in the existing infrastructure: insufficient training resources, fragmented access to compute and data, and steep learning curves for those new to AI. These obstacles highlighted systemic challenges that a national portal could help address. Despite these challenges, participants offered a clear and hopeful vision for a next-generation AI portal. They envisioned a platform that would personalize access based on user roles and experience levels, streamline workflows with explainable AI assistants, and promote responsible data and model reuse through integrated FAIR-aligned tools. Strong themes also emerged around multilingual support and community-building mechanisms that foster collaboration and shared learning across disciplines and institutions. This paper presents key findings and implications from this research study. Rather than prescribing a fixed technical roadmap, offers a community-derived foundation for building a scalable, adaptable, and trustworthy national AI infrastructure that can evolve with emerging needs. BACKGROUND Science gateways have long played a vital role in supporting digital research infrastructure, offering intuitive, web-based interfaces that simplify access to advanced computational tools, curated datasets, and collaborative environments [4]. Known internationally as Virtual Research Environments (VREs), these platforms abstract technical complexity and allow researchers to focus on their domain-specific questions without needing deep expertise in cyberinfrastructure. Despite this potential, AI and ML applications remain underrepresented within gateway ecosystems. A central issue is the lack of awareness among AI/ML developers of how science gateways can reduce usability barriers, streamline access to infrastructure, and enhance reproducibility and transparency. The science gateway community recognizes this gap and has identified a strategic opportunity: to increase the visibility and adoption of gateways in the AI/ML domain while simultaneously leveraging AI/ML to improve gateway services themselves. The integration of AI/ML into gateways supports goals such as software-as-a-service deployment of trained models, explainability and versioning of data and algorithms, automated documentation of AI workflows, and improved user engagement through intelligent digital assistants. Science gateways are also uniquely positioned to facilitate responsible AI by supporting FAIR principles, offering controlled access to sensitive data, and embedding ethical review and transparency mechanisms into platform workflows. This positions them as crucial enablers of reproducible and trustworthy AI research. Furthermore, as AI becomes embedded across disciplines including physics, quantum computing, agriculture, urban planning, and the humanities science gateways can serve as platforms for democratizing access to AI-enabled tools and methods. Gateways such as dREG for genomics [5], Snow Vision for archeological imaging [6], and the Permafrost Discovery Gateway [7] illustrate how ML services embedded in domain-specific gateways can scale knowledge access and enable non-specialists to engage with sophisticated algorithms that would otherwise remain out of reach. This convergence is not limited to user-facing applications. AI also holds promise for transforming gateway infrastructure itself: enhancing usability through natural language processing, improving security through anomaly detection, enabling accessibility audits, and personalizing user experiences [8]. These capabilities, when embedded thoughtfully, support the broader vision of science gateways as not only access points to infrastructure but also intelligent environments that learn and adapt to user communities. The theoretical framework for aligning science gateways and AI/ML must address key challenges: availability of tools, validation and reproducibility of models, transparency and traceability of workflows, and the trustworthiness of outputs. Science gateways offer infrastructure to support each of these, with roles in capturing metadata, providing secure environments, enabling audit trails, and sharing models and results for community validation in both research and educational settings. By grounding development in the real-world needs identified through this research project and leveraging lessons from the growing intersection of AI and science gateways, the NAIRR Portal can become a model for participatory, responsible, and scalable infrastructure to support the national AI research ecosystem now and into the future. PLANNING OF THE SURVEY, FOCUS GROUPS,AND WORKSHOP The research design for this project combined complementary quantitative and qualitative methods to explore user expectations, challenges, and aspirations related to AI infrastructure and research support. A. Survey Design and Implementation The foundation of the community engagement strategy was an online survey developed collaboratively by SGX3 and the Evaluation Services team at the Texas Advanced Computing Center (TACC). Designed around the guiding themes of user needs, portal functionality, AI research activities, training requirements, and responsible AI governance, the survey incorporated both closedand open-ended questions. The survey content was informed by a detailed review of the 2023 NAIRR Task Force Report and pilot-tested with selected stakeholders to refine clarity and relevance before the community survey launch. Launched via Qualtrics from October 2, 2024, to January 2, 2025, the survey collected 1,234 responses, of which 1,057 were valid for analysis. Distribution channels included SGX3 mailing lists, NAIRR user lists, and outreach to PIs from NSFand NIH-funded AI projects. Respondents represented a broad spectrum of institutional affiliations, disciplinary backgrounds, and AI expertise. Key insights from the survey identified critical user priorities, including the need for integrated data management tools, guided resource discovery, and educational content tailored to various roles and expertise levels. These findings established a data-driven baseline for subsequent qualitative engagement. B. Focus Group Structure and Recruitment To delve deeper into the patterns and priorities that surfaced in the survey, SGX3 organized eight focus groups from October 2024 through January 2025, including two held in person at the Gateways 2024 conference. Each twohour session was centered on a defined theme, ranging from workforce development and training to user interface design and research support. Discussion protocols were standardized to ensure consistency, beginning with an overview of the NAIRR initiative, an explanation of participant roles, and targeted prompts derived from the survey responses. Participants were selected through a combination of open calls, ssurvey volunteer responses, and nominations from NAIRR-affiliated organizations. Efforts were made to ensure diverse representation across geographic regions, institution types (including Minority Serving Institutions), and user roles (e.g., researchers, educators, RSEs, and support staff). Notes and transcripts were collected using Otter.ai to capture participant insights with high fidelity. C. Workshop Facilitation Using Design Thinking Building upon insights from the survey and focus groups, SGX3 hosted a two-day, in-person workshop at the San Diego Supercomputer Center in February 2025. Eighty-four participants representing academia, government, and industry engaged in design-thinking exercises structured around curated “problem statements” synthesized from previous engagements. Participants formed twelve working groups, each assigned a user-centered challenge to explore through ideation, prototyping, and user journey mapping. SGX3 had predefined twelve problem statements based on the findings from the survey and focus groups. The workshop started with ideation, prototyping, and lightweight testing. Teams created detailed paper prototypes of potential portal features. OUTCOMES OF THE SURVEY, FOCUS GROUPS,AND WORKSHOP The national survey responses showed that users are actively exploring AI-related activities, but they face persistent challenges in accessing infrastructure and guidance. Respondents emphasized the importance of integrating tools such as Jupyter Notebooks, PyTorch, and RStudio and called for training in machine learning fundamentals and applied workflows. Focus group discussions deepened this understanding, surfacing detailed narratives of workaround-heavy processes, technical bottlenecks, and a lack of institutional support for AI experimentation. Participants called for guided navigation and intelligent recommendation systems to replace the fragmented toolchains currently in use. The design workshop added a layer of specificity and creativity to these insights. Participants created prototypes for AI-powered assistants, FAIR data dashboards, user-role adaptive interfaces, and automated compliance checkers. Educators highlighted the importance of curriculum integration and student-friendly interfaces. These outcomes point toward a vision of a portal that is not only functionally powerful but also intuitive, collaborative, and supportive of broad user empowerment. TECHNICAL DESIGN SUGGESTIONS Across engagements, users envisioned a portal that blends automation with interpretability. Technical suggestions included real-time metadata auditing, consent-aware data sharing, and embedded visual feedback mechanisms to support learning and reproducibility. Participants emphasized the importance of job templates, role-specific dashboards, and onboarding guides that adapt to user experience levels. They proposed modular systems that allow institutions to integrate their environments while benefiting from centralized usability standards. Security was another key topic. Suggestions included federated authentication, granular user permissions, and alerts for policy violations. Participants also stressed the importance of infrastructure that supports versioning, data lineage, and workflow reproducibility. Collectively, these ideas shape a vision for a modular, intelligent, and FAIR-aligned infrastructure that meets researchers where they are. SUGGESTIONS FOR COMMUNITY-ORIENTED FEATURES Community features were considered essential, not optional. Participants called for community “labs” to foster shared experimentation and mentorship, with forums and dashboards that surface active collaborations and new learning materials. Proposed mechanisms for recognition included contributor profiles, citation metrics, digital badges, and spotlight features for outstanding projects or resources. Educators requested tools tailored to instructional use, including assignment templates, class workspaces, and access-controlled student environments. Governance was also a priority. Transparent oversight structures, user advisory councils, and open reporting processes were seen as critical for long-term trust and sustainability. Suggestions reflected a desire for an inclusive and participatory environment that evolves in response to community feedback. OUTLOOK This community-informed effort offers a concrete foundation for advancing the NAIRR Portal in alignment with core NSF priorities. Participants across the national survey, focus groups, and workshop emphasized the need for a modular, user-adaptive portal that supports AI adoption and lowers barriers to entry for all communities. Key recommendations include support for reproducible workflows, integrated training environments, and flexible user interfaces tailored to diverse roles and expertise levels. These features directly contribute to NSF’s goals of promoting open science, transparency, and equitable access to research resources. The findings also underscore the importance of sustained community engagement and governance mechanisms to ensure the portal evolves with user needs. Federated authentication, metadata auditing, and configurable environments were identified as critical technical enablers for scalability, security, and long-term sustainability. The portal’s architecture must also enable scalability without centralization. The focus is to create a shared ecosystem model where local compute and data environments could integrate with centralized NAIRR services through federated authentication, standardized APIs, and reusable templates. This approach supports extensibility while preserving institutional autonomy and resource contributions. SGX3 will continue to support the NAIRR Pilot through the refinement of community-informed requirements, usability testing, and advisory input on governance structures. Future work includes developing functional prototypes of highpriority features such as role-based onboarding and shared community lab environments. By combining technical innovation with persistent community collaboration, SGX3 aims to ensure that the NAIRR Pilot Portal evolves into a sustainable and policy-aligned research infrastructure capable of supporting the next generation of AI-enabled discovery. To fully realize the NAIRR vision, continued investment is needed in both infrastructure and community support. The portal can serve as a critical and complementary cyberinfrastructure component that helps enable accelerate innovation across disciplines using the vast ecosystem of NAIRR resources. ACKNOWLEDGMENT We would like to acknowledge NSF OAC 2231406 (SGX3) and its NAIRR supplements. Furthermore, we acknowledge the survey respondents, the focus groups’ participants and the workshop participants. REFERENCES [1] (2025) Nairr pilot portal. Accessed: 2025-07-09. [Online]. Available: https://nairrpilot.org/ [2] S. Gesing, C. Stirm, M. Zentner, M. Dahan, and L. Hayden, “SGX3: Novel concepts to enhance knowledge and extend the community around science gateways,” in ”Science Gateways 2023 Annual Conference”. Geneva, Switzerland: Zenodo, 2023, pp. 1–5. [Online]. Available: https://zenodo.org/records/10034892 [3] S. Gesing, M. Dahan, C. Stirm, L. Hayden, and J. Baker. Nairr portal report. Zenodo. Creative Commons Attribution 4.0 International (CC BY 4.0). [Online]. Available: https://zenodo.org/records/15792892 [4] P. Calyam, N. Wilkins-Diehr, M. Miller, E. Brookes, R. Arora, A. Chourasia, D. Jennewein, V. Nandigam, M. DrewLaMar, S. Cleveland, G. Newman, S. Wang, I. Zaslavsky, M. Cianfrocco, K. Ellett, D. Tarboton, K. Jeffery, Z. Zhao, J. Gonz´ alez-Aranda, M. Perri, G. Tucker, L. Candela, T. Kiss, and S. Gesing, “Measuring success for a future vision: Defining impact in science gateways/virtual research environments,” vol. 33, no. 19. [Online]. Available: https://onlinelibrary.wiley.com/doi/full/10.1002/cpe.6099 [5] C. G. Danko, S. L. Hyland, L. J. Core, A. L. Martins, C. T. Waters, H. W. Lee, V. G. Cheung, W. L. Kraus, J. T. Lis, and A. Siepel, “Identification of active transcriptional regulatory elements from GRO-seq data,” vol. 12, no. 5, pp. 433–438. [6] C. Wilder, S. T. McDorman, J. Zhou, A. King, Y. Lu, K. Y. Smith, S. Wang, and W. M. J. Simmons, “Snowvision: The promise of algorithmic methods in southeastern archaeological research,” vol. ––, no. ––, p. –, online preprint. [Online]. Available: https://paas.org.pl/wpcontent/uploads/2021/02/co-wilder.pdf [7] Arctic Data Center and Collaborators. Permafrost discovery gateway. National Center for Ecological Analysis and Synthesis; University of Alaska Fairbanks; Alfred Wegener Institute; Woodwell Climate Research Center. Online platform for pan-Arctic permafrost geospatial data and AI-based visualization. [Online]. Available: https://pdg.open.uaf.edu/ [8] S. Gesing, M. Pierce, S. Marru, M. Zentner, K. Huff, S. Bradley, S. B. Cleveland, S. R. Brandt, R. Ramnath, K. Kee, M. Dahan, B. M. V. Mart´ ınez, W. C. Sepulveda, and J. J. S. Mondrag´ on, “Science gateways and ai/ml: How can gateway concepts and solutions meet the needs in data science?” in Critical Infrastructure–Modern Approach and New Developments, M. Santana de Oliveira, B. B. Holl´ o, M. Zafar, E. H. De Aguiar Andrade, and M. M. Radanovi´ c, Eds. IntechOpen. [Online]. Available: https://www.intechopen.com/chapters/86501