reinforcement learning
ABOUT THE ROLE
Join a small applied research team working directly with frontier AI research organizations on challenging post-training problems. This research-first role focuses on reinforcement learning, training data, and experiments that improve model capabilities.
WHAT YOU'LL DO
- Work with customer research teams to turn open-ended goals into concrete environments, datasets, experiments, and deliverables.
- Design and run post-training loops using reinforcement learning methods such as GRPO, preference optimization, and supervised fine-tuning.
- Shape and inspect high-fidelity trajectories and training data, with attention to correctness, edge cases, and quality.
- Build reward functions, rubrics, validators, and evaluation harnesses that resist reward hacking and measure capability gains.
- Analyze post-training signals and experiment results across distributed GPU clusters to determine next steps.
WHAT WE'RE LOOKING FOR
- At least 2 years of experience as an applied researcher or research scientist, with hands-on model training experience.
- Practical reinforcement learning and agent experience beyond standard machine learning engineering work.
- Research experience at a frontier AI lab, an AI research or data lab, or through substantial independent work; alternatively, published research or a released model with strong results on a notable benchmark.
- Experience with Python, PyTorch, FSDP, FastAPI, GRPO, and distributed GPU clusters.
- Experience with post-training methods, training data, reward design, and model evaluation.