Marco Somma

What I do

I am an AI engineer and technical lead building production systems for regulated, high-stakes domains, currently in EU regulatory compliance. My work sits where generative models meet hard correctness requirements: turning stochastic model behaviour into deterministic, auditable checks, and diagnosing why LLM evaluations, provider migrations, and pipelines fail in ways dashboards do not show.

Alongside that I build and write in the open. I am the author of OrKA, an open-source multi-agent orchestration framework, and I run an anti-hype publication on production AI. My longer research runs through complex adaptive systems, symbolic regression, and animal behaviour, all circling one question: how populations of simple parts settle into something coherent without a central authority deciding what is true.

Before AI I trained across art, music, and ethology. I keep aquariums and vivariums, read widely across biology and complexity, and work in Italian, Spanish, English, and Catalan.