RESEARCHER, POST-TRAINING

What you'll do

• Lead research on supervised fine-tuning, reinforcement learning, reward modeling, and preference data.
• Design post-training datasets and evaluation suites; run rigorous experiments and document findings.
• Work with data, infrastructure, and engineering teams to scale reliable training pipelines and address model failure modes.

What they ask for

• Strong track record in post-training research at a frontier-model lab or equivalent.
• 5+ years of hands-on ML research experience.
• Experience with large-scale data curation and preference-data pipelines, and designing evaluations for hard-to-benchmark capabilities.
• Fluency in PyTorch or equivalent, comfort with distributed training, strong statistical instincts, and strong written communication.

Nice to have

• PhD in ML, statistics, CS, or adjacent; or published research at named or comparable venues.
• Experience with reward hacking detection, scaling reward models, RLHF infrastructure, synthetic data, RL math, or open-source post-training infrastructure.

Tools

PyTorch, SFT, RLHF/RLAIF, RLVR, DPO

AI focus

Post-training research for autonomous research-agent models, including supervised fine-tuning, preference data, reinforcement learning from human and AI feedback, reward modeling, evaluation, and training pipeline improvements.