AI architecture diagrams always look clean. Real products rarely are.

In this session, I will share the architecture decisions, experiments and mistakes behind integrating AI capabilities into nVibes, a European social platform designed around real-world human connection.

Instead of presenting a perfect reference architecture, we will look at the uncomfortable questions that appear once AI meets a real product: Which features actually benefit from AI? Where does an agent create unnecessary complexity? How do you keep AI services observable and affordable? How do privacy, identity and trust requirements change the architecture? And when should you simply not use AI at all?

I will walk through decisions that worked, approaches that had to be reconsidered, and the engineering principles that emerged from those lessons.

This is a practitioner session about building under real constraints — architecture, cost, regulation, security and product value — rather than another polished AI demo.