Kubernetes · AI/ML · 7 min read
Make the GPU a first-class citizen in Kubernetes
A practical scheduling model for mixed training and inference workloads, from device discovery to queue policy.
Writing / field notes from connected systems
Thoughts from
the operating layer.
Practical notes about putting models into production, shaping Kubernetes platforms and making infrastructure changes safer to operate.
AI / ML systemsKubernetesNetworkingPlatform security
AI/ML · Delivery · 6 min read
A safer release path for models on Kubernetes
Treat model rollout as a control loop: version the artifact, shift traffic deliberately and let service signals decide what happens next.
Networking · Platform · 6 min read
DNS-as-code needs an IPAM feedback loop
Validated zone files are a start; address ownership, drift detection and rollback turn DNS automation into an operational control.