I build the platforms AI products run on.

Since late 2025 I have worked as an independent platform engineer, and this is my primary engagement. I joined the product team of an AI-security company with a hand-built cloud and a product that could not afford downtime. The first decision was to make the whole GCP estate code: 316 Terraform-managed resources in 12 modules across 9 environments, including 48 build triggers, imported live behind zero-diff gates so nothing changed by accident while the team kept shipping.

Then we made it multi-cloud without storing a single key. Four partner AWS accounts receive our images through identity federation, with the same digest promoted from dev to prod, and one Helm chart now deploys 6 services to 7 GKE and EKS environments.

Security became a system instead of a report: a weekly sweep over every running image, an owner for every finding, and tickets that close themselves when the fix lands. The first run surfaced 12 critical and 388 high findings. And the cloud bill dropped by about seventy percent, through GPU capacity that scales to zero and environments that switch themselves off when nobody is using them.

Alongside my main engagement, a computer-vision company for retail and recycling analytics needed vision that runs where the cameras are, not in a data centre. I built the Jetson Orin pipeline: detection, depth-aware tracking, catalogue recognition, and faces blurred on the device before a single frame leaves it, sustaining about 8 to 9 frames per second at 720p. I wrote the privacy architecture and the impact assessment with the founders, so the guarantee they make to customers is enforced in code, not promised in a policy.

The services run on a five-node edge fleet behind a control plane where devices pull their instructions. There is no path to push code into an edge node, which is exactly the point. Around the edge I built the analytics platform the sales team demos to enterprise prospects, delivered in about ten days, and a reusable deployment scaffold that let us cut the platform over with zero route changes.

Since 2021 I have advised a global engineering talent network on how to tell good engineers from good-looking résumés. I design technical assessments and evaluation frameworks by technology and seniority, review candidates’ code submissions, and help the talent team make screening decisions they can defend. Lately that includes spotting solutions written with external help or generated by AI, and keeping the review standard consistent when that happens.

It keeps me honest about my own work too: reading hundreds of solutions to the same problem is the fastest way to learn what clear code looks like.

The Intel contract turned into three years as tech lead at an Intel-backed smart-city company, leading a team of six on the edge platform behind a 10,000-plus camera deployment across 198 customs sites in Mexico. Each site runs compute, storage and inference on one cluster: ninety days of video on erasure-coded MinIO, clustered failover, more than 300 streams per server with AI analytics on 60 of them.

We shipped a white-label video management system and its SDK, took analytics to three Mexico City municipalities and a retail pilot in Lima, and built the detection pipelines on Intel’s OpenVINO stack. My job was to make a national-scale system boring to operate: automated onboarding, remote management, and a node stack that a field technician could install.

The relationship did not end with the platform. Since 2026 I have built for its digital-signage spin-off: fleet management for 400 Android tablets, and a vendor-neutral proxy that kept the fleet running when the device-management vendor rate-limited us for twelve hours.

Fundación Goodwill brings primary care and rehabilitation to children with disabilities in rural Argentina. Since it started in 2022, 182 volunteers across 24 specialties have seen 504 children, and I have run its systems from the first trip.

I designed and run the platform behind those trips: intake, clinical records, prescriptions, pharmacy, equipment loans and logistics. It is built offline-first, because the places it is used have no signal, and it syncs when the team gets back. AI helps the clinicians where it saves real time: visit transcription, summaries, and a searchable archive of the paper history.