Description
OpenRefinery STEM Research Program
About OpenRefinery
*OpenRefinery is a data infrastructure company focused on improving AI for Science.
*We aggregate real-world data from tools, experts, and operational environments, then refine them into evaluation data, training sets, and reward signals to train AI models in helping with scientific discovery.
The STEM Research Program
The Program is designed to help PhD Candidates and early-career researchers with improving research productivity via better AI tools and training.
What we offer
- $2,400 annual ($200 per month) AI tool grant. Use it for Codex and Claude Code subscriptions, or any other study- or research-related tools.
- AI research resources, tools, and workshops for better science: A hands-on workshop on practical agentic coding workflows, prompt patterns that work for research tasks, and how strong PhDs are using frontier models today. Recorded sessions, live Q&A, and office hours. Based on your own AI usage sessions, we will also provide self-improving AI agents, along with evaluation and benchmarking research results.
- Career development opportunities. Program participants will be first in line for future opportunities with OpenRefinery and our frontier AI lab partners, including but not limited to frontline project work, internships, part-time roles, and recommendation letters.
What we ask
No need to spend extra time on evaluations or assigned tasks. Just use AI tools such as Claude Science, Claude Cowork, ChatGPT Codex, or other AI tools as you normally would in your regular work.
One thing: roughly 5-10 “traces” where AI models have failed to give you the perfect answer, or where they needed your detailed guidance to arrive at the correct answer, or anything you feel would be helpful for understanding that subject.
A "trace" is the raw record of a session with Codex/Claude Code: your prompts, tool calls, model outputs, and files touched. You review what already exists from real work and choose what to share.
How it works
You install an open-source tool (ClawJournal) on your local dev machine. Capture, scanning, PII-redaction, and upload all run automatically with your consent. You can review and approve in a browser-based UI what gets shared with us.
How the data is used
Approved bundles are used by AI labs to evaluate and improve their models. We do not sell them to other vendors, we do not use them for advertising, and we do not publish them publicly. Any use beyond research will require new consent from you.
Current members and backers
PIs, PhD Candidates, postdocs, and other researchers from 30+ institutions, including Stanford, Berkeley, UCSF, MIT, Cornell, University of Wisconsin–Madison, University of Texas, ETH Zurich, National University of Singapore, The University of Sydney and more.
We’re backed by Zetta Ventures (whose founder is the Chair of MIT Board of Trustees), university trustees, and top VCs.
Getting started
If you would like to join, reply to the person who sent you this doc or please email your resume to [email protected]. Once you’re approved, you will receive an onboarding packet with a credit card, user guide for ClawJournal, and a short consent agreement.
