The original title is: "Azure AI Search in 2026, how to build a RAG pipeline"

Short summary

Azure AI Search (now exposed as Foundry IQ in the Azure AI Foundry portal) combines three search technologies: BM25 full-text search, HNSW vector search, and a transformer-based Semantic Ranker. Production RAG pipelines should use hybrid search (BM25 + vector) merged via Reciprocal Rank Fusion, then apply the Semantic Ranker to the top 50 results. This approach solves the problem where pure vector search misses specific terms like error codes or product names that keyword search catches immediately.

• •Azure AI Search combines BM25 full-text, HNSW vector search, and Semantic Ranker cross-encoder re-ranking
• •Hybrid search via Reciprocal Rank Fusion outperforms either vector or keyword search alone for enterprise content
• •Semantic Ranker processes only top 50 results, so initial ranking quality is critical; semantic queries add cost

Generated with AI, which can make mistakes.