Domain LLM

RAG vs Fine-tuning vs Hybrid

Domain data — collection, curation, synthesis

LoRA / QLoRA / DoRA — the production fine-tune

Continued pre-training (CPT) — when worth the spend

Preference tuning — DPO, KTO, GRPO

Domain evals — golden sets, judge, gates

Serving — vLLM with LoRA hot-swap

The on-prem domain assistant