Senior platform and infrastructure engineer moving into AI and ML infrastructure.
I run GPU and compute heavy workloads on Kubernetes, with a focus on reliability and cost.
I have spent more than fifteen years building and running infrastructure across cloud, hybrid, and on-prem environments. My work sits on the systems side: automation, deployment, reliability, and cost control. Over the last few years that has meant production Kubernetes platforms on AWS, infrastructure as code with Terraform, observability with Prometheus and Grafana, and steady work on reducing cloud and compute cost.
I am now focused on AI and ML infrastructure: serving models on GPUs, running them on Kubernetes, and keeping those systems reliable and affordable. This is the infrastructure layer behind AI, not model training or prompt work.
Running model serving and compute heavy workloads on Kubernetes, including GPU scheduling and autoscaling.
Health checks, service level objectives, monitoring, and incident response for production systems.
Measuring and lowering the cost of cloud and GPU compute without giving up reliability.
Repeatable infrastructure as code, CI and CD pipelines, and self service paths for teams.
In progress
A small but production shaped setup that serves an open large language model with vLLM on a GPU, inside Kubernetes. It adds autoscaling, monitoring, and dashboards that report cost per request and per million tokens, and compares the cost of an on-prem GPU against a cloud GPU. The code and a write up will be published when the build is complete.
Kubernetes, vLLM, Terraform, Helm, Prometheus, Grafana, k6
More projects in this area are planned and will appear here as they ship.
Notes and case studies will be posted here. The first will cover the inference platform project above.