Speed Through Structure
Designing a development process around rapid, quantified iterations is
something I think about constantly. Iteration speed needs to be
understood and actively improved. Every iteration needs to move a
metric. And metrics need to be meaningful.
That sounds obvious, but in practice it requires being almost
obsessive about what actually drives iteration speed: understanding
where engineering time goes, what should be more efficient, and taking
the steps to make it so.
Good metrics matter as much as good engineering process. The best
metrics tell you the truth about where the work actually stands.
Without that, the highest iteration velocity will still drive you in
the wrong direction.
Ship With Confidence
Proving that a new system is better, or at least no worse, is one of
the hardest problems in AI development. At Lyft and Toyota, shipping
ML into real-world autonomous vehicles meant every release had to be
proven better, or at least no worse, than what came before. I worked
at the intersection of ML development, validation, operations, and
release to build the processes that got software and ML models onto
public roads.
That experience shaped a practical approach to non-regression and
impact quantification with principles that apply anywhere AI touches
production. This has only become more relevant as we integrate AI
into our development processes, not just our products.
Hands-On, Not Just Advisory
I still build. Not because I have to, but because I want to stay sharp
and because I enjoy creating. Through Fionn Innovation, I've been
building ML pipelines teams can trust: reproducible training, rigorous
benchmarking, faster iteration. Alongside client work, I'm a part-time
founder shipping a consumer computer vision app. Both keep me current
with agentic engineering best practices and pressure-test my own
frameworks.
When I advise on architecture or process, it's grounded in software
I've shipped recently, not slides from years ago.
Drawn to Work That Matters
I'm most energised by teams applying AI to problems worth solving:
climate and energy, mobility, biodiversity, and any work where AI
makes the humans in the room more capable, not optional. Technology is
neutral; where you point it isn't. I enjoy building software that
matters.