The Pulse
Hugging Face Launches ProtoStar to Make Public Science AI-Ready
Hugging Face and university and research partners launched Project ProtoStar to make publicly funded datasets, models and code easier to discover and reuse. Carnegie Mellon University will host the first challenge, testing AI forecasts for
AI.info Team · Written with AI · fact-checked ·
Publicly funded research may be required to reach the public, but that does not mean researchers—or AI systems—can readily use it. Hugging Face’s new Project ProtoStar brings universities, research organizations and open-science groups together to address that gap, with a first challenge asking whether AI can improve climate forecasts for 210 U.S. cities.
The initiative, announced October 29, is built around making research datasets, models and code available in machine-actionable formats, then testing whether others can reuse them. The project’s partners say scientific materials often remain scattered across incompatible systems or lack the documentation, metadata and provenance needed to interpret them.
Public access does not guarantee usable research
“But making research publicly available does not automatically make it usable,” the ProtoStar authors write. The U.S. government invests about $190 billion annually in research and development, the announcement says, while federal policy and agencies including the National Institutes of Health and National Science Foundation expect funded research outputs to be shared publicly. ProtoStar’s premise is that publishing a paper or placing a file online does not, on its own, make the underlying work easy to find, combine or build on.
The initiative’s founding university partners are Carnegie Mellon University, Purdue University, Georgia Institute of Technology, Arizona State University and The George Washington University, working through the Open Forum for AI. Astera Institute, Hugging Face and the Open Research Community Accelerator are also partners. Participating organizations will formalize their involvement through a memorandum of understanding supporting the release of suitable datasets, models and code on open infrastructure.
Hugging Face says its platform already hosts more than 17,000 public datasets in fields such as genomics, protein design, chemistry, weather and climate, Earth observation and astronomy. Including medicine and mathematics brings the figure to nearly 29,000, alongside a comparable number of scientific models. The company also reports that researchers shared more than three times as many new science datasets in the first nine months of 2026 as in the same period of 2025.
CMU’s first challenge tests city climate forecasts
ProtoStar plans a series of open challenges hosted by participating research institutions. The first, hosted by Carnegie Mellon, runs November 2–13 and centers on the Climate Risk Intelligence challenge. Teams will use historical climate records through 2018 to predict observed conditions from 2019 through 2026 across 210 U.S. cities.
Participants will forecast average annual temperature, total annual precipitation, the hottest day of the year and cooling degree days, a measure related to air-conditioning demand. The challenge will compare predictions against straightforward baselines, including historical averages and linear trends. Teams will also submit a short climate-adaptation policy idea based on their analysis.
Forecasts will be scored on a public leaderboard measuring prediction error against those baselines; policy ideas will be considered separately. The organizers say they will publish final forecasting results in early 2027, after complete weather records for 2026 become available. That design makes the challenge an evaluation against observed conditions, rather than a contest built around hypothetical future scenarios.
The project is also a test of research infrastructure
ProtoStar’s challenges are intended to show whether openly released scientific materials can be found, reused, combined and extended, while identifying barriers such as documentation gaps and interoperability problems. The findings, the organizers say, could inform research infrastructure, funding decisions and public policy. Carnegie Mellon’s Open Source Programs Office is already encouraging its researchers to share suitable datasets, AI models and other research outputs on Hugging Face, while continuing to recognize the role of specialized repositories.
Hugging Face cites existing reuse on its platform as evidence of demand: 18 public repositories maintained by NIH’s National Center for Biotechnology Information had received more than 11 million downloads, and NASA IMPACT’s 27 repositories had received about 770,000, as of October 26, 2026. Those figures show activity on the platform; the first challenge will test a more specific question—whether shared data and AI methods can produce better city-level climate predictions than simple statistical baselines.
Researchers and organizations can join
The project is open to academic researchers, students, independent developers and companies, as well as universities, federal research agencies, national laboratories and nonprofits interested in making scientific resources more reusable. ProtoStar invites organizations to contact the Open Research Community Accelerator about participation. Carnegie Mellon’s challenge begins November 2, and the organizers say future challenges will be announced through the Project ProtoStar page on Hugging Face.