Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready

Hugging Face, five U.S. universities, the Astera Institute and the ORCA community network are launching Project ProtoStar, an effort to get publicly funded datasets, models and code released in formats AI systems can actually find and use. Participating institutions sign a memorandum committing to release suitable artifacts on open infrastructure.

The gap is real. Federal policy increasingly requires research to be shared, but shared is not the same as usable. Data sits in legacy formats, scattered systems, and without the documentation that lets anyone else build on it. The first open challenge, hosted by Carnegie Mellon, runs November 2 to 13.

This is the unglamorous side of AI for science, and we think it is the side that matters most. Breakthrough claims get headlines. Clean, documented, reusable data is what actually lets a grad student or a small lab build on someone else's work. Hugging Face's own number makes the point: 90% of dataset downloads on its platform go to fully documented datasets.

The demand signal is there. Hugging Face says researchers shared more than three times as many new science datasets in the first nine months of 2026 as in the same stretch of 2025, and NIH's NCBI repositories on the platform have passed 11 million downloads. When public institutions publish well, people use it.

I like the design of the first challenge. Teams use climate records from 210 U.S. cities through 2018 to predict what actually happened from 2019 to 2026, and get judged against simple baselines like historical averages and linear trends. That is honest. It asks whether AI beats the boring method, not whether it looks impressive.

The downside to name is concentration. A lot of public science flowing onto one company's hub, however open, creates a dependency, and the post itself stresses that domain-specific repositories still matter. We'd also want to see the policy-idea component of the challenge handled carefully, since a forecast leaderboard is not the same as sound adaptation advice. Still, this is the kind of AI infrastructure work we want more of.

Founding partners include Carnegie Mellon, Purdue, Georgia Tech, Arizona State and George Washington University through the Open Forum for AI, plus the Astera Institute, Hugging Face as infrastructure partner and ORCA as community partner. Hugging Face is one of ten generalist repositories NIH lists for sharing research data.

The U.S. government invests roughly $190 billion a year in R&D, and agencies like NIH and NSF have tightened public-access and data-sharing requirements.