Generative AI involves not only generating outputs but also ensuring they are useful, safe, and reliable. Evaluation is crucial to detect issues like inaccuracies, hallucinations, or style mismatches. When base models fall short, customization through methods like fine-tuning or RAG becomes necessary.