JOB-1916

A growing AI startup partners with frontier model labs on their toughest post-training challenges and is adding applied researchers to a small existing team as customer demand rises. The position is research-first, prizing deep understanding of reinforcement learning and training data over pure engineering throughput.

Your focus

  • Collaborate with the researchers at customer labs to convert fuzzy goals into specific deliverables, experiments, datasets and environments
  • Design and run post-training loops using supervised fine-tuning, preference optimization or RL methods such as GRPO, and work out which signal truly improves a model
  • Curate and examine detailed trajectories and datasets with an eye for correctness, corner cases and judgment
  • Construct evaluation harnesses, validators, rubrics and reward functions that withstand reward hacking and reveal real capability gains
  • Read experiment outputs from large GPU clusters and choose the next experiment

What you bring

  • At least two years as an applied researcher or research scientist
  • Practical RL experience gained inside a frontier lab, a younger model lab or AI data lab, or through a substantial solo project with results to show
  • Published research or a released model that performed well on a notable benchmark
  • Production post-training experience
  • A PhD involving RL is a plus

Hybrid in New York, NY or Seattle. Visa transfers and new sponsorships are both possible.