Design and implement supervised fine-tuning pipelines, create benchmarks, analyze and improve production LLMs, run experiments and hyperparameter searches, build Python tooling for training/evaluation, and document research results.