Researcher, Post-Training

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 team based in San Francisco.

Responsibilities

• Lead post-training research including SFT, RLHF/RLAIF, RLVR, DPO, and reward modeling
• Design and curate data for post-training from sourcing to quality assessment
• Build and maintain evaluation suites to measure model capabilities
• Run rigorous experiments with controls, ablations, and statistical significance
• Scale data pipelines and infrastructure for training
• Identify and characterize failure modes like reward hacking and distribution drift

Requirements

• 5+ years of hands-on ML research experience
• Strong track record of post-training research (SFT, RL, reward modeling) at a frontier-model lab or equivalent
• Proficiency in PyTorch or equivalent for distributed training
• Experience with large-scale data curation and preference-data pipelines
• Experience designing evaluation suites for complex capabilities
• Strong statistical instincts and written communication skills

Preferred Qualifications

• PhD in ML, statistics, CS, or adjacent fields
• Published research at NeurIPS, ICML, ICLR, COLM, or RLC
• Experience with reward hacking detection or RLHF infrastructure
• Experience with synthetic data generation
• Background in RL math (policy gradients, importance sampling, off-policy methods)
• Open-source contributions to post-training infrastructure