NAIRR Pilot Newsletter November 2026
This monthly newsletter provides updates on the National AI Research Resource (NAIRR) Pilot Program, highlighting resources, opportunities for researchers, and achievements from the NAIRR community in advancing AI across various scientific domains.
In This Issue
Getting Started with NAIRR: Helpful Presentations
Research Highlight: Transforming Safety on U.S. Construction Sites
A team from Michigan Technological University is using NAIRR resources to explore how generative AI can interpret job-site video streams, reason through OSHA safety regulations and detect construction job site hazards in real time.
Construction is one of the most dangerous industries in the United States, with hazards such as electrocution and struck-by injuries (getting hit by moving objects) claiming thousands of lives each year. Despite its national economic significance, the field remains one of the least digitized, creating major barriers to developing effective AI tools that can monitor job sites and prevent accidents.
At Michigan Technological University, Assistant Professor Bo Xiao and his interdisciplinary research team are working to change that. Their project, supported by NAIRR, explores how generative AI can interpret job-site video streams, reason through OSHA safety regulations and detect jobsite hazards in real time, with the ultimate goal of protecting workers before accidents occur.
Traditional construction safety monitoring relies heavily on manual oversight, intermittent inspections, and delayed reporting – practices that offer limited protection in fast-changing environments. Previous attempts to automate safety detection using computer vision faced significant limitations. Most construction datasets contain only a few hundred images, preventing models from generalizing across different job site conditions.
Models trained to detect struck-by hazards often fail when evaluating electrocution risks. Computing constraints can also make real-time analysis nearly impossible where processing a single video feed could take minutes or crash local servers. These gaps motivated Xiao’s team to explore generative AI, promising stronger reasoning capabilities, and improved generalizability. However, the required computing power was far beyond what the team had available.
Through the NAIRR, the team used Azure cloud-based GPU resources to test advanced multimodal and generative AI models. These resources allowed the team to run high-resolution video pipelines without hardware failures, test state-of-the-art image captioning and safety reasoning models, use Azure Cognitive Search to map OSHA regulations to visual scenes, build a secure web-based hazard detection prototype, and support hands-on training for undergraduate and graduate researchers. Previously, model training could take days – or crash entirely – but with NAIRR Pilot resources, large-language models can be immediately applied for new construction scenarios, enabling rapid experimentation.
Dr. Xiao recalls the pivotal moment: “The ‘aha’ moment came when a video stream that once crippled our server ran smoothly through a generative model on NAIRR's cloud resources. That was when we realized real-time construction safety AI could actually become a reality.”
Looking ahead, Dr. Xiao sees this project as a starting point for exploring the full potential of generative AI in construction safety. By combining computer vision, language processing, and IoT technologies, his team hopes to better understand how AI can prevent hazards in real time and provide actionable safety guidance on job sites. With NAIRR resources enabling faster experimentation and more sophisticated models, the project offers opportunities for broader collaboration with the AI community. Ultimately, Dr. Xiao aims to demonstrate how advanced computing can enhance safety management, empower researchers and engineers, and spark new approaches to protecting workers in one of the world’s most challenging industries. “Our hope,” he says, “is that these tools will one day make construction a fundamentally safer profession.”
For more information on this project, watch the NAIRR Pilot Researcher Showcase #3.
Publications
Hou, X., Walavalkar, A., & Xiao, B. (2026). A generative AI-based construction safety assistant using retrieval-augmented generation. In Proceedings of the 43rd International Symposium on Automation and Robotics in Construction, 2329–36. IAARC, 2026. https://doi.org/10.22260/ISARC2026/0297
Funding and Support Acknowledgements
The work in this highlight has been supported by the NAIRR Pilot ID #240362.
NAIRR Deep Partnerships
The NAIRR Deep Partnerships Opportunity enables researchers and teams to propose a collaboration with one of the industry partners on projects of mutual interest.
Current deep partnership opportunities include:
Lexset Seahaven: Offering researchers the opportunity to make custom visual synthetic datasets for Computer Vision model training and evaluation. Specifically, researchers can gain supported web-based and API access to Lexset's Seahaven synthetic data generation platform.
NSDS Secure Compute Environment: Allows approved researchers to securely access, link, and analyze data for approved projects within a secure data enclave. The environment also enables testing and use of state-of-the-art privacy-enhancing technologies and AI/machine learning techniques.
OpenMined: A unique opportunity for researchers to unlock the potential of non-public datasets through privacy-preserving technologies. This Deep Partnership enables research consortia to collaborate on sensitive data while maintaining data privacy and security, effectively providing access to 1000x more data for AI research.
TACC Horizon: Supports machine learning and GPU-enabled applications with a system based on NVIDIA Grace, Vera, and Blackwell architecture. Horizon is intended for parallel applications scalable to thousands of cores, machine learning workflows, and simulation applications that utilize GPUs. Normal batch queues will enable users to run simulations up to 48 hours.
Featured Training and Events
OSPool User Training - Keep it Running: How to Checkpoint your OSPool Jobs
Tuesday, November 17 2:30 PM - 4:30 PM EST
If your job needs to run for days, this training is for you! Learn how to run workflows with checkpointing capabilities on the OSPool. In this training, we will walk through a hands-on example using GROMACS, a popular molecular dynamics simulation software, but no molecular dynamics knowledge is needed to participate.
Topics covered:
- When you should checkpoint your jobs
- How to implement exit-driven checkpointing in your scripts and submit files
- Common mistakes and troubleshooting tips
ACCESS & NAIRR Pilot Office Hours
Tuesday, November 10 3:00 PM - 4:00 PM EST
Wednesday, November 25 12:00 PM - 1:00 PM EST
Office Hours are hosted by SCIPE Cyberinfrastructure Professionals (CIPs) and representatives from NSF’s ACCESS program, and are regularly attended by members of the NAIRR Pilot operations teams and Pilot partners. We are happy to answer questions related to proposal submission, active NAIRR Pilot awards, ACCESS MATCH submissions and ACCESS resource usage and support.
CIPs in the SCIPE program come from a diverse set of scientific disciplines and institutions, which may not be the institution hosting your allocated resource. Providing a detailed description of your issue or help request at the time of registration will give us an opportunity to prepare for your consultation. If your issue isn't resolved during Office Hours, SCIPE CIPs or ACCESS staff will connect you with the appropriate Resource Provider team.
This newsletter provides information about resources made available by the NAIRR Pilot. The NAIRR Pilot is a joint effort led by the National Science Foundation in collaboration with other U.S. federal agencies.
Follow NAIRR Pilot on YouTube
This newsletter is brought to you by Internet2, with support from the National Science Foundation under Grant No. 2436057. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.