{"id":4879,"date":"2026-03-15T19:01:01","date_gmt":"2026-03-15T18:01:01","guid":{"rendered":"https:\/\/gokhan-gokalp.com\/?p=4879"},"modified":"2026-03-15T19:11:35","modified_gmt":"2026-03-15T18:11:35","slug":"building-an-ai-agent-in-net-deterministic-routing-and-intelligent-search-with-microsoft-agent-framework","status":"publish","type":"post","link":"https:\/\/gokhan-gokalp.com\/tr\/building-an-ai-agent-in-net-deterministic-routing-and-intelligent-search-with-microsoft-agent-framework\/","title":{"rendered":"Building an AI Agent in .NET: Deterministic Routing and Intelligent Search with Microsoft Agent Framework"},"content":{"rendered":"\n

Nowadays wherever I look, everyone talks about AI<\/em> coding agents, agentic systems or LLM<\/em> powered applications and so on. It has been almost a year that I have been working on different domains where we integrate LLMs<\/em> into our applications, as well as getting my hands dirtier by integrating AI assisted coding practices into our SDLC<\/em> lifecycles.<\/p>\n\n\n\n

To be honest, after spending some time with these, I really liked the new way of working and how these tools make things simpler compared to what and how we used to build. Lately, I have been also playing with Microsoft’s<\/em><\/strong> newAgent Framework<\/strong><\/em>\u00a0after spending some time with Semantic Kernel<\/em> before, and I must say, Microsoft<\/em> plays the game very well! I really liked how much it simplifies the experience compared to Semantic Kernel <\/em>although since it is still quite new and evolving, the documentation and community resources are not as mature yet.<\/p>\n\n\n\n

What I also liked is that <\/em>Microsoft Agent Framework <\/em>(MAF<\/em>) does not just give us an agent runner. It comes with building blocks we actually need in real projects. Model clients for chat completions, session management for state, context providers for memory, middleware for intercepting and observing agent actions and MCP <\/em>clients for tool integration. We are not starting from scratch or gluing things together ourselves, the foundation is already there.<\/p>\n\n\n\n

In this article, I will try to show how we can build a simple but practical AI<\/em> agent<\/strong> using MAF. <\/em>The agent we will build is a product search assistant that uses deterministic routing<\/strong> as the backbone and an intelligent search capability powered by Azure AI Search<\/strong><\/em> as the RAG<\/em> store where MAF’s<\/em> agent automatically handles the reasoning<\/strong> loop internally.<\/p>\n\n\n\n

Let’s Get Our Hands Dirtier<\/h1>\n\n\n\n

Before getting into the code, there are two Azure<\/em> services we need to have ready. The first one is Azure OpenAI<\/strong><\/em> with a model deployed. I used “gpt-4.1-mini<\/em>” for this article which is lightweight and more than capable for intent classification<\/strong> and tool-based reasoning<\/strong>. If you don’t have one yet, you can follow this<\/a><\/em> guide to create a resource and deploy a model and keep keys, endpoints and deployed model names somewhere handy.<\/p>\n\n\n

\n
\"\"<\/a><\/figure>\n<\/div>\n\n\n

<\/p>\n\n\n\n

The second one is Azure AI Search<\/strong><\/em>, which will serve as our RAG<\/em> store, the place where our agent will look up products. You can also follow this<\/a><\/em> guide to create a free tier instance and again keep key and endpoint information.<\/p>\n\n\n\n

So, we are going to build a minimal product search assistant exposed as a REST API<\/em>. A user sends a natural language query, the agent classifies the intent, routes to the
right executor and the search agent retrieves and reasons over matching products from Azure AI Search<\/em>. The interesting part is how we wire all of this together with deterministic<\/em><\/strong> routing<\/em><\/strong> for control and an agentic reasoning <\/strong><\/em>loop for the intelligent part\u00a0which MAF<\/em> handles for us automatically under the hood.<\/p>\n\n\n\n

First, let’s create a minimal .NET API<\/em> project by running the following command.<\/p>\n\n\n\n

<\/circle><\/circle><\/circle><\/g><\/svg><\/span>