AI and LLM application evaluation is the practice of systematically assessing the quality, safety, and performance of large language model applications across development and production environments. Unlike traditional software testing, LLM evaluation requires measuring subjective qualities like relevance, coherence, and factual accuracy alongside objective metrics like latency and cost—making it both an engineering and human-centered discipline. Modern evaluation spans multiple layers: offline benchmarking with datasets, online monitoring with real user interactions, and specialized frameworks for RAG systems, agents, and multi-step workflows. The key insight: what you don't measure, you can't improve—systematic evaluation transforms LLM applications from unpredictable experiments into reliable production systems.