Building AI-powered applications with .NET
AI presents an enormous opportunity to make your applications more powerful and productive. It can automate many existing business processes, and can let your users zip through their tasks at much greater speed while remaining in control. Intelligent features can include semantic search, data extraction, anomaly detection, translation, summarization, sentiment scoring, autocompletion, classification and workflow automation, and of course Q&A chatbots that work with enterprise data.
For .NET developers, there’s a whole raft of new libraries, services, models, and concepts to get your teeth into. This workshop aims to cut through the complexity and focus on what you really need to know to get started with AI programming on .NET. It will emphasize foundational topics in .NET+AI, including the new Microsoft.Extensions.AI standards, then go onto the next level of understanding of real-world production AI systems.
You’ll learn:
- Core concepts of AI for app developers
- Embeddings, semantic search, vector databases, vector indexing
- Large language models, chat, function calling, structured output, middleware pipelines
- Q&A chat/RAG: Ingestion/chunking, retrieval-augmented-generation, evaluation
- Vision and multimodality, automating business processes
- Prompt engineering and prompt injection attacks/defenses
- Realtime AI (just released by OpenAI)
- Agentic systems with AutoGen
- Advanced RAG patterns with agents, implementing CRAG
- Lessons learned from AI in production and at scale
In each case we’ll have explanations/demos plus guided exercises in C#, some being open-ended for more advanced exploration. We hope what you’ll get out of it is a broad understanding of current-day AI+.NET app development, and a sense that you’ve experienced most of the main pieces
Prerequisites
Familiarity with C# and typical .NET application patterns, such as dependency injection
You don’t need to know web programming with ASP.NET Core or Blazor in any significant detail, though one or two exercises will use them. Mostly we’ll work in console apps to preserve focus on AI.
What to bring
You’ll need a Windows/Linux/macOS laptop with:
- .NET 9 – whatever the latest preview is
- An editor – either Visual Studio, VS Code, or Rider
- Docker
- Ollama (and while you’re on a fast network at home, run “ollama pull llama3.1” because that’s a 4.7GB download)
Optional
If at all possible, also have an OpenAI platform subscription or Azure OpenAIdeployment. It’s not strictly required since you can just use Ollama, but OpenAI/AzureOpenAI will be much faster – especially if your laptop doesn’t have a good GPU. And you’ll learn more if you can compare small models on Ollama with the big ones like GPT 4o.
Steve Sanderson is a developer/architect on the .NET team at Microsoft. He's enthusiastic about helping the developer community move forwards, currently with a focus on web and AI.
Diego is working as a Principal Software Engineer at Microsoft. Over the past 20 years he is been working in lots of different industries using .NET technologies in video games, finance and robotics. With a researcher background he still gives presentations and lectures in Universities around Europe.