You will lead end-to-end post-training for agentic models, managing everything from SFT warm starts to large-scale on-policy RL runs. Your work involves optimizing model performance, reliability, and efficiency while ensuring training runs are reproducible across various compute environments.
You will design and build synthetic data environments and task generators to train and evaluate large language models. You will also be responsible for creating rigorous testing and evaluation frameworks to identify model shortcuts and ensure high-quality performance.
You will build reward models and judges to ensure LLM reliability and honesty during training. You will also conduct alignment experiments and develop safety evaluations that are effective in real-world scenarios.
