More than twenty years ago I started building computer vision algorithms to detect lung cancer in CT scans. That work pulled me deep into AI/ML engineering, medical imaging, and eventually into leading teams that ship production software in one of the hardest domains there is: healthcare, life sciences, and medtech.

Along the way I've architected cloud-native ML platforms, built RAG pipelines and multi-agent workflows, led GPU-accelerated inference deployments on AWS, and taken products through FDA 510(k) clearance. Before moving to enterprise, I spent nearly two decades as a Principal Investigator leading multidisciplinary R&D teams across computer vision, medical imaging, and clinical AI — work that taught me how to bridge the gap between research ambition and production reality.

What shipping production AI in regulated, safety-critical environments teaches you is that the hard part is never the model. It's the infrastructure, the evaluation rigor, the governance, the observability, and the ability to work across engineering, clinical, and business teams simultaneously. Most AI initiatives stall because they treat these as afterthoughts.

That's exactly the problem I'm now focused on at enterprise scale. I work with large organizations to move AI out of the experiment phase and into their actual operating model. Designing agent architectures with retrieval, orchestration, routing, and observability built in from day one. Building abstraction layers across AI providers so teams aren't locked into a single vendor. Rethinking workflows from first principles around what AI makes possible, rather than wrapping AI around processes that were designed for a pre-AI world.