Head of Engineering.
Still commits code.

My way into technology began over a decade and a half ago and was a long detour. I started coding at 14 — side-projects, small experiments, some freelance websites, a Visual Basic order system for a friend's restaurant, a learning app for other students while I was working as tutor at university. The usual self-taught patchwork. But for years it stayed a clandestine hobby running alongside what I considered my real studies: language and cognition.

I studied Linguistics and Cognition, first in Bari and then in Göttingen, spending most of my twenties on questions about how human language works, synchronically and diachronically. It was in that academic context, of all places, that I built my first neural networks, inspired by Paul Smolensky's work on Gradient Symbolic Computation. Modelling phonological constraints with vector-space representations turned out to be the bridge.

I find the symmetry quietly satisfying. The languages I now spend my days working with are programming languages, but they are still languages. Most of what I learned about how meaning is built in human grammar transfers surprisingly well to how systems acquire meaning at scale.

The weight of my role has shifted toward people. There is real satisfaction in growing engineers into Tech Leads or Team Leads, and watching a system run more confidently every quarter. I've also done the harder side of that work: probationary endings, mutual exits, performance turnarounds. Leaders who can't make those calls kindly are unfit for senior roles; leaders who can make them without empathy are unfit for any role.

The technical side never really left. I've spent the last year authoring the company-wide AI tooling strategy, designing the supporting infrastructure, and building multiple agent systems to increase team efficiency.

On AI in general: engineering leaders cannot afford to delegate AI judgement to anyone else on the team. Productive skepticism — the discipline of using these tools deeply enough to know where they break — is part of the job now. That skepticism has an operational edge too: AI may suggest the lines, but accountability belongs to whoever pushes the button. Never commit what you cannot explain or defend in a postmortem — fifteen words that have shaped how our teams work with AI in practice.