AI systems in production: containers, CI/CD and Azure burst

What it does. The platform that keeps these products running every day: they deploy themselves, roll back in minutes and push heavy AI jobs to the cloud on demand.

Problem. AI systems that only work on a laptop are demos; they need deploys, rollback and capacity for heavy jobs.

My role. Platform design and operation.

How it works. Nine own systems in containers with health checks, CI/CD on GitHub Actions, images pinned by digest and tested blue/green rollback. Heavy render jobs burst to ephemeral Azure Container Instances from one central queue, with a local fallback.

Result. Nine production systems running in containers; 8 of 9 deploy themselves on merge to main, with rollback.

Stack. Docker · GitHub Actions · GHCR · Azure Container Instances · Linux