LLM fine-tuning pipeline for persona-based dialogue. Train models to embody documentary characters with LoRA. Includes training, evaluation, and GGUF export.