Deliver production-ready AI search on unstructured data with RAG | ODSP925

Microsoft Developer explains how to take a RAG prototype and turn it into a production-ready AI search experience over unstructured data, including a simple end-to-end pipeline and practical patterns for scaling relevance and performance.

Overview

The session focuses on moving from proof of concept to production-ready retrieval-augmented generation (RAG) for AI search over unstructured data.

It covers:

• How to build a simple RAG pipeline end-to-end
• Common architectural complexity points when operationalizing RAG
• Patterns for scaling and improving retrieval quality, including:
  • Agentic RAG
  • Graph-based retrieval
  • Entity recognition
• How to choose approaches based on performance, relevance, and maintainability

Demo highlights

• Building a financial dashboard with .NET and Blazor
• Using a C# SDK and Blazor to create custom interfaces
• Displaying data with Telerik UI for Blazor

Session chapters

• 0:00 - Why retrieval augmented generation is important today
• 00:01:59 - Definition of context augmented generation and examples from YouTube Gemini
• 00:04:26 - Data ingestion and vector embeddings in RAG
• 00:05:27 - Complexity of end-to-end RAG architecture
• 00:07:59 - Rapid search experience creation with HTML widget builder
• 00:08:32 - Demo: Building a financial dashboard with .NET and Blazor
• 00:11:33 - Using the C# SDK and Blazor for custom interfaces
• 00:12:54 - Displaying data with Telerik UI for Blazor
• 00:13:55 - Closing and accessing additional resources

Event context

This is a Microsoft Build 2026 on-demand session (ODSP925).