Led research on robustness, evaluation, interpretability, and post-training alignment for language models. Worked on data quality, supervised fine-tuning, reinforcement learning from human feedback, and the Zephyr model.