Hugging Face announced Project ProtoStar on Thursday, an initiative to make publicly funded scientific research easier to find, reuse and build on with AI.
The pitch is that grant-backed research should ship in forms a machine can act on, and that the argument is proven by putting it to work. Hence the challenges: each one takes a body of openly released science and asks what can be built on top of it. Five universities anchor the effort, Carnegie Mellon, Purdue, Georgia Tech, Arizona State and George Washington, gathered under the Open Forum for AI. The Astera Institute is involved, Hugging Face supplies infrastructure and the Open Research Community Accelerator acts as community partner. Members formalize the commitment in a memorandum of understanding.
The gap it targets is real. Washington spends roughly $190B a year on research and development, and agencies like the NIH and NSF increasingly require the results and underlying data to be shared. Hugging Face is one of ten generalist repositories the NIH lists. But publishing is not the same as usable. Data often sits in legacy formats, scattered across incompatible systems, with no documentation of where it came from.
Demand is already visible. The platform hosts more than 17,000 public datasets in core scientific domains, approaching 29,000 when medicine and mathematics are counted. NIH’s biotech repository has drawn over 11 million downloads, and 90 percent of dataset downloads come from fully documented ones.
The first challenge, hosted at Carnegie Mellon, opens November 2 and asks teams to predict urban climate conditions from historical weather records.
