Research Scientist (Full Time/SFO, CA/Onsite/USA)

Job Title: Research Scientist

Location: San Francisco, CA (Onsite)

Job Type: Full Time/Permanent

Visa: Full Time/USC/GC/Etc..

The company: a days-old AI agents lab, founders from Tesla AI, Anthropic, DeepMind, NVIDIA Research and Jane Street.

Requirements

Exceptional ML research and engineering ability

Deep experience in one or more of: LLM post-training, RL, agents, reasoning, long-horizon systems, memory/context, computer use, evals, or human-agent interaction

Strong experimental judgment; builds complete systems, not isolated prototypes

Moves fluidly between research ideas, large-scale experimentation, and production systems

A record of significant research ownership: an important model, agent system, benchmark, paper, open-source project, or major research bet

High agency and ambition; comfortable with uncertain directions where the right approach isn't known

Intense, high-ownership; thrives at a days-old company

Background

~3–6 years; frontier-lab research/engineering experience valued (open to exceptional outliers; senior leads welcome above)

"Spike" backgrounds a plus: olympiad (IOI/IMO), quant, world-class competitive achievement

Nice-to-Have

Work directly on RL post-training, agents, long-horizon systems, computer-use, memory/context, evals, or human-agent interaction

Strong publication / open-source / benchmark record in these areas

Environments / training-data-pipeline / eval-harness building experience

Who Will Thrive Here

A researcher who wants to own long-horizon agent bets end to end, training through deployment

Someone who builds complete systems (environments, pipelines, evals), not just papers

High-agency, comfortable in the fog of uncertain research

Excited to shape a research agenda at a days-old company

Mission-aligned with long-horizon autonomous agents

Process ends with a paid 2 to 3 day research work trial in SF, expenses covered. Fully in office. Visa sponsorship available; a strong publication or open-source record makes the O-1 realistic.