Project ProtoStar launches to make public science AI-ready

Hugging Face and research partners launched Project ProtoStar on October 29, aiming to make publicly funded scientific datasets, models and code easier for researchers and AI systems to reuse. Carnegie Mellon University will host the first open challenge from November 2 through November 13.

Participating organizations formalize their involvement through non-binding, non-exclusive memoranda of understanding. The agreements cover suitable publicly funded datasets, models, code and documentation for release on open infrastructure in AI-ready formats—machine-readable files with metadata and instructions that help others discover and reuse them. The ProtoStar organization page says each organization retains control over which materials are appropriate to release and how they are shared.

The founding group includes Carnegie Mellon, Purdue University, Georgia Tech, Arizona State University, The George Washington University and Astera Institute. The universities participate through the Open Forum for AI. Hugging Face is the infrastructure partner, and the Open Research Community Accelerator participates through its Fostering Innovation Network. The project’s MOU index lists six signed agreements, including one with Georgia Tech Research Corporation.

ProtoStar plans a series of challenges in which participating institutions release scientific resources for researchers and developers to reuse. The goals are to produce concrete examples of reuse and uncover technical, documentation and interoperability barriers.

In the inaugural Climate Risk Intelligence challenge, teams will use historical weather records for 210 U.S. cities through 2018. They must forecast 2019–2025 annual average temperature, total precipitation, hottest-day temperature and cooling degree days, then propose a climate-adaptation policy idea. Submissions require a public Hugging Face dataset with 5,880 forecast rows, a README explaining the method, policy proposal and feedback on the ecosystem, and the code used to generate the forecasts.

The challenge draws on a 2019 CMU-authored Journal of Climate study of long-term city-level U.S. temperature and precipitation records. One detail remains inconsistent: the launch post says teams will predict conditions through 2026, while the ProtoStar page and detailed CMU rules specify 2019–2025. The detailed rules set the submission deadline at 11:59 p.m. Eastern Time on November 13.