ABOUT THE COMPANY
We're building autonomous research agents for recursive self-improvement (multi-agent systems that propose, run, and analyze machine learning experiments). We're a small, well-funded team that works onsite in San Francisco.
ABOUT THE ROLE
As a Researcher on our team, you'll design experiments and develop methods that drive how our autonomous research agents make decisions. You'll work across the full ML research stack (problem formulation, method design, experimentation, analysis, write-up) and you'll do it on problems that don't always have established benchmarks because we're inventing the workloads.
The work is open-ended and concrete at the same time. Open-ended because the research problems are constantly evolving and we don’t prescribe approaches. Concrete because the research questions are motivated by real-world applications. Open-ended because we don't have prescribed research directions; concrete because every experiment ties to something the agents will actually do. You'll have real autonomy (and the corresponding responsibility for choosing well).
WHAT YOU'LL DO
Identify research questions that, when answered, would meaningfully change what our agents are capable of
aDesign and run experiments end-to-end (from problem framing through method design, infrastructure, evaluation, and write-up)
Develop new methods spanning RL, LLMs, agentic systems, multi-agent coordination, search, evaluation, or wherever the problem leads
Work closely with engineers to take the most promising methods from research code into production
Read deeply across the literature; bring useful work from outside in
Help shape how the team picks problems
WHAT WE'RE LOOKING FOR
Strong track record of ML research at the frontier: RL, LLMs, agentic ML, multi-agent systems, evaluation, or adjacent
5+ years of hands-on research experience in industry or academia
Comfortable designing experiments and running them at scale, not just proposing them
Strong written communication: you can summarize your research findings into actionable insights for next steps
Fluent in PyTorch, Jax or equivalent; comfortable working with large-scale training infrastructure
Bias toward shipping research rather than handing it off
Comfortable with ambiguity: many of our problems don't have a known right answer, and navigating that uncertainty is core to the role.
Published research at NeurIPS, ICML, ICLR, COLM, RLC, or comparable venues
NICE TO HAVE
PhD in ML, statistics, computer science, or adjacent
Open-source contributions to ML research infrastructure
Experience with agentic systems, tool use, long-horizon planning, or multi-agent coordination
THIS ROLE IS PROBABLY NOT FOR YOU IF
You want to focus on one specific benchmark and watch the metric tick up (our problems are broader and shift)
You prefer more pure research that never touches a production system
You'd rather work alone than share research taste openly with a small team