Fine-tuning and model adaptation
Decide when to use prompting, RAG, fine-tuning, or a combination. Prepare instruction and preference datasets; clean, deduplicate, split, and version them. Run parameter-efficient fine-tuning using LoRA/QLoRA-style approaches and evaluate whether adaptation improves task performance
Hands-onBuild a domain-adaptation pipeline for a narrow task such as internal policy Q&A, support-ticket classification, or controlled report generation. Include dataset preparation, training-run configuration, evaluation before and after adaptation, cost tracking, and a decision memo recommending fine-tuning versus retrieval-based alternatives.
