Hi, I’m Geert Baeke. I’m chief technologist at Inity in Belgium. I build and teach cloud-native and AI solutions on Azure, and this blog is where I write down what I learn along the way.

Most posts start the same way: I want to understand how something works, so I build a small project, break it a few times, and then write it up. Most of the code is on GitHub. Expect code, screenshots and honest caveats.

What I work on

A large part of my work happens before anyone writes code. I sit with customers to understand what they need, and turn those requirements into an architecture and a design that fit: which services to use and how they connect, and how security and cost hold up once it runs in production. After that I help teams build it, and I teach the courses and workshops that go with it.

These days most of those projects involve AI: agents that work with a company’s own data and systems, and the Azure platform they run on. The technology I use most:

  • Agents in code: LangGraph, Microsoft Agent Framework and Foundry Hosted Agents, with RAG and multi-agent setups on a wide range of models
  • Low-code agents in the Microsoft stack: Copilot Studio and Microsoft 365
  • Agent platforms and governance: gateways such as Azure API Management, and Agent 365
  • Azure architecture in general, with implementations on AKS, Container Apps, Container Apps Sandboxes, Functions and Microsoft Foundry, deployed with Bicep

If you’d like to work with me on a project for your business, drop me a mail at geert@baeke.info. You can also find out more about what we do at Inity.

Recent projects

A few things I’ve built recently. Several have a post on this blog.

🌙 dottie

Persistent personal agents that sleep until needed: Deep Agents, a PostgreSQL wiki as memory and Azure Container Apps Sandboxes.

🔀 model-router

Classifies prompt complexity on-device and routes each prompt to a cheap or a strong model via agentgateway.

☸️ kagent-substrate

A local proof of concept: kagent 1.0 on Agent Substrate in kind, with an in-cluster Ollama.

📡 mcp-events

Exploring the draft MCP Events spec with a FastMCP server that ChatGPT can subscribe to.

🧠 neurosymbolic-hr-demo

An LLM extracts facts from leave requests, and clingo (ASP) applies the HR policy.

🎙️ realtime-webrtc

OpenAI realtime audio with WebRTC, in the browser.

Find me elsewhere

If you spot an error in a post or have a question, mail me or send a message on LinkedIn.