Retrieval Augmented Generation (RAG) pipelines enhance AI systems by integrating document retrieval with language generation, aiming to provide accurate and grounded responses. A RAG system combines a Retriever, which searches a knowledge base, with a Generator that formulates answers from retrieved data, reducing inaccuracies often referred to as hallucinations. Evaluating these systems is crucial, particularly in sensitive domains like healthcare and finance, to ensure accountability and reliability. The evaluation focuses on retrieval quality, generation faithfulness, and end-to-end effectiveness, employing frameworks like three-stage evaluation, inputs-to-insights, and feedback loops to systematically assess performance. Optimization strategies such as semantic chunking, retriever fine-tuning, reranking, and prompt scaffolding are employed to enhance relevance, coherence, and factual consistency of the output. These strategies and evaluations are essential for developing trustworthy, scalable RAG systems that perform effectively in real-world applications.