Azure AI Search Proxy Job Support — From Vector Search to Agentic Retrieval

Agentic retrieval uses an LLM to plan and run retrieval: a knowledge base (the orchestrator, formerly the "Knowledge Agent" concept) exposes a retrieve action over knowledge sources (indexed or remote). At low/medium reasoning effort it decomposes a query into parallel subqueries, semantically reranks each, and merges results — optionally synthesizing an answer. Core features are GA in the 2026-04-01 REST API; the full set including answer synthesis and non-web LLM knowledge sources is in the 2026-05-01-preview API, and the portal exposes it as preview.

Keyword, vector, and hybrid search; the semantic ranker; integrated vectorization and push indexing; indexers, skillsets, and data sources (Blob, ADLS, SQL, Cosmos DB, and more); knowledge bases, knowledge sources, and the retrieve action for agentic retrieval; answer synthesis; scoring profiles, filters, facets, and analyzers; and security-trimming and RBAC. We also connect Search to Azure OpenAI RAG, Foundry IQ, and Foundry agents.

Yes — this is the most common request. We diagnose why results are irrelevant: chunking and embedding choices, vector-dimension mismatch, missing hybrid search or semantic ranking, bad filters, stale indexes, or skillset failures. We tune the pipeline and evaluate retrieval quality so answers actually improve.

Yes. Indexer and skillset failures, throttling during ingestion, change-tracking and deletion detection, integrated-vectorization errors, and document-cracking issues are all in scope. We get ingestion reliable and incremental.

Both. We firefight live retrieval and indexer incidents and prepare you for Azure AI Search and RAG system-design interviews. See our Azure AI Search production and interview support pages.