AI track: AI engineering with .NET

Who it is for

For .NET developers who want to bring AI features into real applications – secure, measurable and with capped costs.

Prerequisites

K2 or the placement test. K3 helps for the module on operations.

Afterward, you can

• Understand language models, tokens and costs and choose the right model
• Build AI features with Microsoft.Extensions.AI, with structured answers
• Build vector search and RAG that cites its sources
• Secure AI features against prompt injection
• Build agents with Microsoft Agent Framework, with approvals by people
• Write MCP servers and clients in .NET
• Evaluate AI features with test sets, also in the build
• Run AI features: tracing, budgets and safe changes of model and prompt

Contents

1. T1

   Foundations

   how language models work, tokens and costs, choosing a model, responsible AI, the labeling duties of the EU AI Act in outline

   15 h
2. T2

   Microsoft.Extensions.AI

   IChatClient, streaming, structured answers with a strict JSON schema, middleware for metering, caching and logging

   20 h
3. T3

   Prompts as code

   system prompts, examples, versioning, the prompt cache

   15 h
4. T4

   Embeddings and vector search

   Microsoft.Extensions.VectorData, Azure AI Search or PostgreSQL with pgvector, chunking

   20 h
5. T5

   RAG

   grounding, citing sources, declining without a source, hybrid search

   20 h
6. T6

   Security

   direct and indirect prompt injection, content filters, Prompt Shields, foreign text fenced in as data, hard rules in code rather than in the prompt

   15 h
7. T7

   Agents with Microsoft Agent Framework

   tools and function calling, workflows, a human in the loop (approvals), memory, several agents

   25 h
8. T8

   MCP

   server and client in .NET, permissions for tools

   10 h
9. T9

   Evaluation

   test sets, Microsoft.Extensions.AI.Evaluation (quality, safety, reports), LLM-as-a-judge with care, evaluation in CI

   20 h
10. T10

    Operations

    GenAIOps touching on MLOps: Foundry projects and deployments, tracing with OpenTelemetry, budgets and caps, watching quality and drift, rolling out model and prompt changes safely; classic ML in overview (ML.NET, Azure Machine Learning)

    20 h
11. T11

    Final project

    AI in the Kladde

    20 h

About 200 hours in all (indicative).

Lessons

You see the text and video of every lesson here on the site after buying the course; the titles and goals are already shown. A self-study course: you work through the lessons in order and check your exercises yourself against the solutions.

The lessons are only in German for now.

The lessons stand under their modules, in the order of the course. Click a module to open or close its lessons.

Final project: the Kladde

Certifications

An offer, never a requirement: these Microsoft exams fit this stage. You take them at Microsoft; we are not a Microsoft training partner.

• AI-901 Azure AI Fundamentals

  The entry point to the AI certifications, centered on Microsoft Foundry; expects a basic knowledge of Python syntax.

• AI-103 Azure AI Apps and Agents Developer Associate

  Expects Python skills.

• AI-500 Multi-Agent AI Solutions Expert

  Requires AI-103 and covers Agent Framework and MCP; expects Python skills.

• AI-300 Machine Learning Operations Engineer Associate

  Assumes a background in data science; expects Python skills.

Microsoft's AI exams expect Python skills; the AI track teaches C#.

Keep learning with Microsoft Learn

Microsoft's learning paths and documentation that go with this course.

• Develop .NET apps with AI features
• Build an AI chat app with .NET
• Invoke .NET functions using an AI model
• Build a .NET AI vector search app
• Microsoft Agent Framework
• The Microsoft.Extensions.AI.Evaluation libraries
• Get started with AI agent development on Azure
• Develop AI agents on Azure