Hugging Face launches Project ProtoStar to make public science AI-ready

Hugging Face has announced Project ProtoStar, an initiative to make publicly funded scientific datasets, models and code easier to find, reuse and build on, particularly for AI-driven research. It brings together universities, research organizations, industry and the open science community to release scientific artifacts in AI-ready, machine-actionable formats, and to show their value through a series of open challenges.

The founding partners are Carnegie Mellon University, Purdue University, Georgia Institute of Technology, Arizona State University and The George Washington University, all working through the Open Forum for AI (OFAI) as the university coalition partner. The Astera Institute is also a founding partner, Hugging Face is the infrastructure partner, and the Open Research Community Accelerator (ORCA), through its Fostering Innovation Network, is the community partner. Participating organizations formalize their involvement through a memorandum of understanding that supports releasing suitable datasets, models and code in AI-ready formats on open infrastructure. Organizations that want to join can write to protostar@orcaopen.org.

The post frames the problem as one of usability. The U.S. federal government invests approximately $190 billion annually in research and development, and agencies such as the NIH and NSF have strengthened public-access requirements and require researchers to plan for sharing underlying data. Hugging Face is among the ten generalist repositories listed by NIH for sharing research data. But making research public does not automatically make it usable: datasets, models and code are often scattered across incompatible systems, kept in static or legacy formats, or released without the documentation, metadata and provenance needed to reuse them. That makes it harder for researchers to build on earlier work, and harder still for AI systems to discover, interpret and combine resources across disciplines. ProtoStar aims to help close the gap through systematic sharing in AI-ready formats, backed by open infrastructure and practical publishing guidance. It builds on the National Cancer Institute's AI Data-Readiness Challenge and Hugging Face's guide to preparing public-sector data for AI.

Hugging Face offers usage figures as evidence of demand. It hosts more than 17,000 public datasets across core scientific domains (genomics, protein design, chemistry, weather and climate, Earth observation and astronomy), a number that approaches 29,000 when medicine and mathematics are included, with a comparable number of scientific models. 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, and growth stays substantial even after excluding the largest individual uploader. The 18 public repositories maintained by NIH's National Center for Biotechnology Information on Hugging Face have collectively received more than 11 million downloads, and NASA IMPACT's 27 repositories have received approximately 770,000 downloads as of October 26, 2026. The post also says 90% of dataset downloads on Hugging Face are from fully documented datasets. It points to HuggingFaceBio releases such as Carbon-A and Carbon-X as examples of researchers building on open genomic data and releasing the results openly.

As an early example of institutional commitment, CMU's Open Source Programs Office is encouraging its researchers to share suitable datasets, AI models and other outputs on Hugging Face, while recognizing that domain-specific repositories remain important.

The central question for ProtoStar is whether publicly funded scientific datasets, models and code, released in AI-ready formats, can be readily discovered, reused, combined and extended to speed up scientific discovery and public benefit. To test it, the project will run a series of open challenges, each hosted by a participating institution and built on resources that institution has released openly.

The first, the Climate Risk Intelligence challenge hosted by Carnegie Mellon, runs November 2 to 13, 2026. Participants test whether AI can improve predictions of urban climate conditions using historical weather data from 210 U.S. cities, building on CMU research. Teams use climate records through 2018 to predict actual conditions from 2019 to 2026, so models are judged against observed weather rather than hypothetical future scenarios. They forecast four indicators relevant to urban planning: average annual temperature, total annual precipitation, the hottest day of the year, and cooling degree days, a measure of demand for air conditioning. Predictions are compared against simple statistical baselines, including historical averages and linear trends, to see whether AI offers meaningful accuracy gains.

Participants also propose a short climate adaptation policy idea based on what they learned, for example on preparing cities for extreme heat, improving energy efficiency or managing changing precipitation. Teams submit a Hugging Face dataset with their outputs. Forecasts are judged on a public leaderboard of prediction errors against baselines, and policy ideas receive separate consideration. Final forecasting results will be published in early 2027, once complete 2026 weather records are available.

Other participating institutions will host their own challenges one after another, each highlighting different publicly funded resources. Future challenges will be announced through the Project ProtoStar page on Hugging Face. The post invites academic researchers, students, independent developers and companies to take part, and universities, federal research agencies, national laboratories and nonprofits to join as partners.

Key facts

Project ProtoStar is a Hugging Face-announced initiative to release publicly funded datasets, models and code in AI-ready, machine-actionable formats; partners sign a memorandum of understanding.
Founding partners: Carnegie Mellon, Purdue, Georgia Tech, Arizona State and George Washington University (via OFAI), the Astera Institute, Hugging Face as infrastructure partner, and ORCA as community partner.
The first challenge, hosted by CMU, runs November 2 to 13, 2026: forecast four climate indicators for 210 U.S. cities for 2019 to 2026 using records through 2018, scored against historical-average and linear-trend baselines.
Hugging Face reports more than 17,000 public science datasets (approaching 29,000 with medicine and mathematics) and more than three times as many new science datasets in the first nine months of 2026 as in the same period of 2025.
Final forecasting results are due in early 2027; other institutions will host later challenges.

Why it matters

Governments pay for a great deal of research and increasingly expect results to be shared; the post cites about $190 billion a year in U.S. federal R&D spending and stronger NIH and NSF public-access rules. ProtoStar targets the step after sharing: data that is public but scattered, in legacy formats or missing documentation and provenance is hard for people, and harder for AI systems, to use. The initiative pairs that aim with concrete open challenges rather than only a call for better practice.

Who it affects

Research institutions that fund or hold datasets, models and code are the main audience: universities, federal research agencies, national laboratories and nonprofits are invited to join. Academic researchers, students, independent developers and companies can take part in the challenges. Researchers at partner universities may see institutional policies encouraging sharing on open infrastructure, as at CMU's Open Source Programs Office.

How to use it

To join the first challenge, see the Climate Risk Intelligence page on Hugging Face; it runs November 2 to 13, 2026. Teams forecast average annual temperature, total annual precipitation, the hottest day of the year and cooling degree days for 210 U.S. cities for 2019 to 2026, add a short climate adaptation policy idea, and submit a Hugging Face dataset with their outputs. Organizations that want to become partners can contact protostar@orcaopen.org. Future challenges will be announced on the ProtoStar page on Hugging Face.

How solid is it

This is an announcement by Hugging Face, which is also the infrastructure partner, so the usage figures (dataset counts, downloads, growth) come from the company's own Hub data and are presented as evidence of demand. The post names no individual authors or spokespeople. The challenge design is described in detail, but the post gives no prize amounts, no baseline error figures and no expected accuracy gains. It does not say how many organizations have signed the memorandum beyond the named founding partners, or what the Astera Institute contributes.

Risks and caveats

The initiative is at launch stage: no results exist yet, and final forecasting results are not due until early 2027. Dates and hosts for challenges after the CMU one are not given. The headline growth figure is a ratio of new dataset counts, and the post gives no absolute number of new datasets. The post itself notes that domain-specific repositories remain important, and that making artifacts AI-ready is only a first step.

“But making research publicly available does not automatically make it usable.”

— Hugging Face, Project ProtoStar announcement