Skip to content
Neural Daily

ResearchNews

Project ProtoStar makes public science AI-ready

🇭🇺 Olvasd el magyarul →

Hugging Face, US universities and nonprofit partners launched Project ProtoStar to release publicly funded datasets, models and code in AI-ready, reusable formats. The first open challenge starts November 2.

By: Mikola Máté Patrik · · 2 min read

Project ProtoStar announcement graphic from Hugging Face
Image: Hugging Face · Press kit · Official thumbnail from the Hugging Face blog · Source

What is Project ProtoStar?

Hugging Face announced the initiative on October 29 (Hugging Face). The goal is to publish publicly funded scientific datasets, models and code in documented, machine-actionable, AI-ready formats on open infrastructure. Participating organizations sign a memorandum of understanding.

  • Founding universities through the Open Forum for AI (OFAI): Carnegie Mellon, Purdue, Georgia Tech, Arizona State and George Washington University.
  • Other partners: the Astera Institute, Hugging Face as infrastructure partner and ORCA as community partner.

Why it is needed

The US federal government spends about $190 billion a year on R&D, and NIH and NSF increasingly expect data to be shared. The post argues that public data is not automatically usable: scattered systems, legacy formats and missing metadata make reuse hard, especially for AI systems.

By Hugging Face's own count, the Hub hosts more than 17,000 public science datasets, close to 29,000 with medicine and mathematics included. Researchers shared three times as many new science datasets in the first nine months of 2026 as a year earlier, and 90% of dataset downloads go to fully documented datasets.

First challenge: Climate Risk Intelligence

Hosted by Carnegie Mellon from November 2 to 13, the challenge asks teams to use weather records through 2018 for 210 US cities to predict actual 2019 to 2026 conditions: annual mean temperature, annual precipitation, the hottest day of the year and cooling degree days. Predictions are scored against simple statistical baselines on a public leaderboard, submissions are Hugging Face datasets, and teams also propose a short climate adaptation policy idea. Final results come in early 2027.

What it means for you

If you work with scientific data, expect more documented, ready-to-load datasets on the Hub and less data wrangling. The challenge is a clean test of whether an AI model actually beats a simple baseline.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles