Microsoft Build 2025: Azure AI Search
If you use Azure AI Search for your RAG applications, you are going to be excited about the new features that were just announced at Microsoft Build 2025.
If you want a video that provides a good overview, I’d recommend watching Pablo Castro’s talk Knowledge Retrieval: RAG for enterprise agents with Azure AI Search, he did a great job highlighting most of the new features.
Agentic Retrieval (new feature)
From the documentation site: “… agentic retrieval is a new parallel query processing architecture that incorporates user conversation history and Azure OpenAI models to plan, retrieve and synthesize queries for improved results. It produces high-quality grounding data for custom chat and generative AI solutions that include agents.”
This blog post walks through highlights: Introducing agentic retrieval in Azure AI Search and discusses how it works, introducing this image as the high level view:
This blog post gets into the details of the classic search vs. the agentic search and highlights its benefits: Up to 40% better relevance for complex queries with new agentic retrieval engine. This image shows the highlight (taken from Pablo’s presentation):
Here are some code examples:
- Quickstart: Agentic retrieval in Azure AI Search - https://github.com/Azure-Samples/azure-search-python-samples/blob/main/Quickstart-Agentic-Retrieval/quickstart-agentic-retrieval.ipynb
- A demo site that uses it - https://github.com/Azure-Samples/azure-search-openai-demo/
- Agentic retrieval using Azure AI Search and Azure AI Agent Service - https://github.com/Azure-Samples/azure-search-python-samples/blob/main/agentic-retrieval-pipeline-example/agent-example.ipynb
- This looks like an interesting walkthrough: Build an agent-to-agent retrieval solution using Azure AI Search
Multimodal Search (multiple new features)
From the documentation site: “Diagrams, scanned forms, screenshots, and infographics often contain the decisive details that make or break an answer. Multimodal search helps close the gap by integrating visual content into the same retrieval pipeline as text. This approach reduces the likelihood that your AI agent or RAG application might overlook important images and enables your users to trace every provided answer back to its original source.”
This is the functionality that will get a description (verbalization) of what an image is and save that description with the text.
This blog post discusses how multimodal search is different than just OCR and why it matters: From diagrams to dialogue: Introducing new multimodal functionality in Azure AI Search
The features that are rolled into this are:
- New parameters for the Document Layout Skill in the 2025-05-01-preview API version that support image offset metadata.
- A new Gen AI prompt skill – when you use the wizard it uses this skill to call OpenAI to get the verbalization of an image.
- New wizard features that will build out the skillset for you to wire up all the pieces
Here are some resources:
- Quickstart: Search for multimodal content in the Azure portal
- Github repo: auzre-ai-search-multimodal-sample – which is a multimodal RAG demo app
Azure Logic Apps integration
This allows for more data sources to be used to import data into Azure AI Search and makes the workflows and connectors in Logic Apps available in that importing.
More information here: Use an Azure Logic Apps workflow for indexer-based indexing in Azure AI Search
Ability to add access control for Entra ID
This solves for ACLs on files and can have Azure AI Search have that capability of not returning results if a user doesn’t have access to the underlying document/chunk/etc. Currently files from Azure Data Lake Storage can carry their ACL’s into Azure AI Search and they are working on other sources to follow.
If you have a comment, please message me @haleyjason on twitter/X.