4.2. Azure AI Search for Retrieval
💡 First Principle: Azure AI Search is the retrieval engine, and its power is combining three different matching strategies — keyword (lexical), vector (semantic), and a re-ranking layer (semantic ranker) — so the system catches both exact terms and conceptual meaning. Knowing which strategy catches which kind of query is the core skill.
Why care: the exam refresh added Azure AI Search as the canonical grounding store, and the recurring question is "which search mode for this requirement?" Pure vector misses exact identifiers (product codes, IDs); pure keyword misses paraphrase and concept. The production-correct answer is almost always hybrid, often with semantic ranking on top.
⚠️ Common Misconception: "Vector search and keyword search return the same results, just ranked differently." They surface genuinely different candidates — keyword matches exact tokens, vector matches meaning. A query for "SKU-4471 return policy" needs keyword to catch the code and vector to catch the policy concept; neither alone suffices.
