I build software with AI for problems nobody had digitized yet.

An AI agent that covers last-minute shift absences over WhatsApp — and the four bugs I only found because I stopped trusting my own test suite.

Inside a mission-critical Angular/.NET platform, a recurring internal task was still done by hand-editing YAML. Here's the spike, the case I made to the client, and what actually made it to production.

Building AI chatbots in .NET often raises the question: which OpenAI model should you use? In this post, inspired by Anton Martyniuk’s guide, we explore how to run multiple models like gpt-3.5-turbo and gpt-4 in parallel, store their responses separately using Neon Postgres branching, and compare results efficiently. It’s a scalable and clean way to experiment with AI without mixing data or overcomplicating your backend.