Microsoft's C# blueprint for ChatGPT over private data

What it does This is an opinionated sample—not a library—showing how to build a ChatGPT-style interface over your own documents using Azure’s AI stack. A Blazor WebAssembly frontend talks to an ASP.NET Core Minimal API backend that orchestrates Azure AI Search for vector retrieval and Azure OpenAI’s gpt-4o-mini for generation. It ships with fake “Contoso Electronics” HR data so you can deploy and test immediately.

The interesting bit Instead of hand-rolling prompt orchestration, it leans on Semantic Kernel to manage the conversation flow between the retriever and the language model. The UI also exposes tuning knobs directly to the user—unusual for a demo—letting you tweak retrieval behavior without redeploying.

Key highlights

• Voice chat, text chat, and Q&A modes with inline citations and source tracking
• Vector search through Azure AI Search over documents stored in Blob Storage
• Optional GPT-4V support, plus toggles for authentication and Application Insights
• Pre-baked Bicep infrastructure and Azure Developer CLI (azd) templates for one-shot deployment
• Includes cost estimation guidance and warns that Container Apps and AI Search incur monthly charges even at rest

Caveats

• This is a sample app and infrastructure template, not a reusable library; expect to fork and modify heavily
• The README is upfront about costs: Container Apps and AI Search carry monthly charges, and Document Intelligence bills per page processed
• Deployment is region-sensitive; the default gpt-4o-mini model currently restricts hosting to East US

Verdict .NET developers who need an official, batteries-included RAG reference architecture for Azure. Not for those seeking a framework-agnostic or budget-free local experiment.