While supervised fine-tuning is allowing LLMs to succeed in narrow contexts, it requires high-quality, domain-specific ...