A Retrieval-Augmented Generation (RAG) pipeline is a crucial component of many AI systems, and evaluating its performance is essential to ensure the quality of the final output. RAG evaluation metrics are used to assess the retriever and generator components separately, focusing on common failure modes within each stage of the pipeline. The five key industry-standard metrics for RAG evaluation are answer relevancy, faithfulness, contextual relevancy, contextual recall, and contextual precision. These metrics help identify issues such as hallucinations, poor chunking strategies, weak reranking logic, and suboptimal Top-K settings. A custom metric called G-Eval is also used to evaluate the generator's performance in specific tasks. RAG evaluation can be performed end-to-end or at a component level, using tools like DeepEval, which provides a comprehensive platform for evaluating LLM applications on the cloud. By incorporating RAG evaluation into CI/CD pipelines, developers can ensure the quality of their AI systems and safeguard against potential issues.