Whether it's for your team or just for you, a private ChatGPT that answers from your own documents has become essential. Getting one shouldn't require exposing your documents, prompts and API keys to the public internet. In this workshop we build a complete retrieval-augmented chat application in .NET 10 and deploy it with Bicep into a private Azure network with no public data plane and no API keys.

The solution includes a reusable C# library for Azure OpenAI, an Event Grid and Azure Functions ingestion pipeline, Azure SQL native vector search, and an ASP.NET Core chat app with citations, image generation and voice. Everything is deployed with Bicep and GitHub Actions. The end result is a fully functional RAG system running on a private network in Azure.

In this workshop, you will learn to:

Build a reusable, testable Azure OpenAI library in C# for chat, embeddings, images and speech
Implement RAG with Event Grid, Azure Functions, EF Core and Azure SQL native vector search
Deploy an AI workload with Bicep into a private network with no public endpoints, using managed identity and OIDC pipelines

In this hands-on workshop, you'll start by deploying a locked-down network in Azure using Bicep, then implement chat orchestration, saved conversations, PDF and image generation, voice, and a document knowledge engine that uses embeddings and Azure SQL vector search instead of Azure AI Search.

Prerequisites:
An Azure subscription where you have Owner rights, or Contributor plus User Access Administrator, so you can deploy resources and assign RBAC roles to managed identities. You also need a GitHub account to run the deployment workflows.