OpenRefinery
OpenRefinery partners with academic scientists to capture the undocumented research process—failed experiments, dead ends, and judgment calls—and turn it into AI training resources. The company helps researchers build evaluations, benchmarks, and datasets that AI labs license, with researchers retaining ownership and sharing revenue. It works with scientists at institutions including Stanford, MIT, Caltech, and ETH Zürich.
- Artificial Intelligence
- Software Only
- Biotechnology
Funding
Founders
Founder details are not available yet.
Product
Problem
Scientific AI models are trained primarily on published papers, which omit the actual research process—the failed experiments, dead ends, and human judgment that shape every result. This missing evidence limits AI's ability to truly understand and accelerate scientific discovery, leaving valuable insights locked in the undocumented work of researchers.
Solution
OpenRefinery.ai gathers the hidden evidence of research practice—questions, dead ends, and judgment calls—at new scale and turns it into AI that accelerates discovery. The company works directly with researchers at leading universities to identify cases where AI has the most to learn, and covers the costs of AI tools for researchers who choose to share their work. Researchers define what a correct result requires, and OpenRefinery's engineers help build evaluations, benchmarks, and training datasets that AI labs can license. By capturing what papers never show, OpenRefinery.ai trains AI systems on the full research process rather than just its published outcomes.
Target Audience
Primary users are academic researchers at leading universities who want to contribute their research process data and gain access to AI tools that accelerate their scientific discovery work.
Features
- AI training grounded in undocumented research workflows, including failed experiments and decision-making processes
- Direct partnerships with researchers at Stanford, MIT, Caltech, ETH Zürich, UPenn, Berkeley, NUS, and other leading universities
- Cost coverage for AI tools provided to researchers who opt to share their research process data
- Engineers help researchers turn methods and data into evaluations, benchmarks, or training datasets for licensing
- Researchers retain ownership of their work and share revenue from licensed resources
- Planned research discussions, benchmark workshops, and social gatherings to facilitate collaboration among researchers