Most AI compliance failures are not model failures or legal failures. They are architecture failures. Built wrong from the start, before anyone asked what the EU AI Act would require.

My current focus, working under Aevoxis Solutions (a project of SMartDe eG) is EU AI Act compliance engineering for agentic and generative AI systems — specifically Articles 9, 12, 13, 14, 17, and 50, translated from regulatory obligation into production architecture decisions. 

I have been identifying these kinds of structural gaps before the industry names them. A decade ago that meant a patented VR headset designed around human ergonomics when the industry was still ignoring them. Now it means AI systems designed around EU AI Act obligations before most engineering teams have read the regulation.

The pattern is the same. The domain has changed.