Context
The company runs its workloads on AWS EKS, including GPU nodes for heavier jobs, and needed clearer cost visibility and more deliberate cluster operations.
Decisions
- Took on AWS and EKS operations directly, closer to a senior infrastructure engineer with a mandate than an advisor.
- Provisioned GPU nodes for the workloads that needed them, and checked whether that capacity was used or sitting idle on the bill.
- Ran it on a weekly cadence, with concrete numbers and small permanent improvements instead of a rewrite nobody had time for.
Result
A recurring operating rhythm focused on cost visibility and workload optimization, without a disruptive rebuild.