Overview
Revised the 2014 container platform in 2018, moving it onto EKS, ArgoCD, Helm, and GitHub Actions. The shift was from infrastructure that had to be operated to infrastructure that reconciles itself.
Context
Great Data Lake licensed Lumate’s technology, which meant inheriting the platform that ran it. The 2014 framework worked, but four years is a long time in container tooling: ECS-era patterns had been overtaken by Kubernetes, and deployment still depended on people running things in the right order.
The question was not whether the old platform functioned. It was what it cost to keep operating, and whether a small team could carry it while doing the actual data-brokerage work.
The System
The move to EKS brought the platform onto Kubernetes, with Helm packaging services so that a deployment became a versioned artifact rather than a sequence of steps. ArgoCD introduced GitOps: the desired state lives in a repository, and the cluster continuously reconciles toward it. GitHub Actions replaced the older build path.
The practical effect is that deployment stops being an event someone performs and becomes a property the system maintains. A drifted cluster corrects itself. A rollback is a revert.
Outcomes
The platform became something a small team could operate alongside other work, which was the real constraint at Great Data Lake — the same lesson that shows up across this portfolio, where the durable win is usually reducing what has to be actively run rather than adding capability.
Patterns Worth Reusing
Inheriting a working system is a real decision, not a formality. The 2014 framework was not broken, and rebuilding it would have been indefensible. What justified the work was operating cost: the platform had to be maintainable by fewer people than built it.
Declarative delivery pays off most where attention is scarcest. GitOps matters less for large teams with dedicated platform engineers than for small ones, where the alternative is knowledge living in one person’s head.