About Scrimdata
Scrimdata builds frontier RL environments from real company data. We license the operational history of real companies, run it through our own anonymization pipeline, and turn it into multi-step, multi-tool training environments: tasks, rubrics, verifiers, and reference trajectories, graded against what actually happened. We're an early-stage, remote-first team selling to two sides of one market: AI labs and enterprise agent teams who train on our environments, and data partners who supply the raw material. Small team, hard problems, real customers from day one.
About the fellowship
The Scrimdata Research Fellowship is a paid, remote, fixed-term program, typically 3 to 6 months and flexible depending on the project, for researchers doing serious work on RL environments. It's built for people early in their research career or between roles: PhD students on a break from coursework, postdocs looking for applied experience, and industry researchers who want to work on a hard, underexplored problem without switching jobs. Fellows work directly with the core team on open questions in environment construction, and the expectation is that the work is good enough to publish, not just useful enough to ship internally.
What fellows work on
- Task mining: finding the real multi-step, multi-tool work buried in an anonymized digital twin and turning it into a well-specified environment
- Verifier and reward design: building programmatic checks that score agent trajectories reliably against what actually happened, not just what looks plausible
- Difficulty calibration: building pass@k tiers and curricula that separate genuine capability from lucky guesses
- Contamination studies: measuring how much of a public benchmark's signal has already leaked into model training, and what that means for the environments we build
- Transfer experiments: testing whether training on real-work messiness (the ambiguous thread, the stale ticket, the conflicting version) produces agents that hold up better than ones trained on clean, synthetic tasks
- Anonymization research: studying how far identity can be stripped from a company's data while keeping the structure, timelines, and dependencies that make the data worth training on
Who should apply
- PhD students, postdocs, or industry researchers with a genuine research background in ML, RL, NLP, or a closely related field
- People with a publication record, strong preprints, or a portfolio of technical work they can point to, even if it's not formally published yet
- Researchers who are comfortable with messy, real-world data and don't need a clean benchmark to get started
- People who want to work on a specific, well-scoped question and see it through to a result, rather than rotate across unrelated tasks
- Strong written communicators who can turn a research question into a clear writeup others can build on
- Anyone currently between roles, finishing a degree, or looking for focused applied research experience before committing to a full-time position
What you get
A paid fellowship for the full term. Direct, hands-on mentorship from the core team, not a rotation through a training program. You publish the work under your own name. Fellows who do strong work are the first people we turn to when we're hiring full-time, and a number of fellowships convert directly into offers.
How to apply
Email careers@scrimdata.com with your resume and a short research statement, or a pointer to prior work: a paper, a preprint, a GitHub repo, whatever best shows how you think. No forms, no links required beyond that.