Luma Health cuts EKS costs by 40% while freeing up thousands of engineering hours a year

Manual Kubernetes optimization was costing more than it saved

Luma Health's finance team holds every department to a strict cloud budget tied to the company's gross margin targets. For Caio’s SRE team, that meant a relentless monthly cycle. “We would allocate at least 20 hours a month to work on cost optimization in the SRE team. We were constantly reviewing metrics, understanding where costs were increasing, and then adding follow-up action items to mitigate those.” shares Caio.

The process was painstaking. Engineers reviewed Datadog metrics, evaluated CPU and memory utilization across hundreds of pods, and manually adjusted resource requests and limits for deployments and DaemonSets. They experimented with general-purpose AI tools like Claude and ChatGPT to speed up the analysis, but the results were inconsistent.

Kubernetes resource tuning demands awareness of traffic seasonality, workload-specific patterns, and the cascading effects of a single change across hundreds of pods. A modest reduction in CPU allocation could starve a service. An overly conservative memory limit could trigger out-of-memory crashes and pod restarts. Caio's team caught most of these risks, but the margin for error was thin, and the time investment was unsustainable.

Caio found himself in a recurring conversation: justify the spend increase to finance, or find equivalent savings elsewhere to offset the investment. "Everyone is very interested in making sure the customer is happy," Caio explains, "and the customer is not happy if a page takes 10 seconds to load." The SRE team needed a way to optimize Kubernetes costs with confidence so they could reinvest those savings into the infrastructure that patients depend on.

Why Luma Health chose PerfectScale

Caio discovered PerfectScale by DoiT while evaluating Kubernetes optimization tools. He shares: “PerfectScale was a fit right from the start. I didn't even continue the trials with the other tools.”

PerfectScale gave Caio's team automated, environment-aware optimization that their manual process never could. The key: optimization policies tailored to different risk tolerances. "We have specific profiles that we use. For lower environments, we use a more aggressive cost optimization approach. In production, we have a more balanced approach.” Explains Caio, “Based on the output and metrics, we can easily apply those changes to our platform."

In non-production environments, PerfectScale automation runs continuously, rightsizing pods to minimize waste without manual intervention. In production, the balanced policy weighs cost savings against the reliability guarantees that healthcare demands. The distinction matters: a platform serving 100 million patients cannot tolerate the same aggressiveness in production that works in a staging cluster with minimal load.

PerfectScale's recommendations draw on continuous workload analysis rather than simple averages, accounting for the traffic seasonality and usage spikes that tripped up the team's previous process. Caio notes that the overwhelming majority of recommendations proved safe to apply to production, and in lower environments the system applies changes automatically.

"The trust factor was something that was very important. More than 90% of the recommendations were accurate. It was safe to apply to production" shares Caio.

PerfectScale’s support team provided hands-on guidance throughout the deployment, helping tune policies and troubleshoot edge cases as Caio's team scaled automation across clusters.

40% EKS Cost Reduction and +1,700 Engineering Hours Reclaimed

The impact was immediate. In the first optimization cycle, PerfectScale identified a 40% reduction in EKS-specific costs, accounting for roughly 15% of Luma Health's total AWS spend.

The time savings proved equally transformative. Caio's SRE team saw a 90% reduction in time spent on Kubernetes rightsizing, freeing engineers across three time zones to focus on the work that actually moves the platform forward.

Perhaps most importantly, PerfectScale gave Caio a credible answer to the recurring budget conversation with Luma Health's finance team. When the cache layer required additional investment to maintain application performance, the SRE team could point to verified Kubernetes savings as an offset. The conversation shifted from defending costs to reallocating savings strategically. AWS continued to provide the reliable, scalable compute foundation through Amazon EKS, while PerfectScale by DoiT ensured every dollar spent on that foundation worked as hard as possible.

"One thing that PerfectScale helped us a lot was to balance this cost. We need to increase our cache layer, but is there anything we can save on the other end? The first time we applied the recommendations, we hit approximately 40% a year in EKS savings. That definitely helps towards conversations where we need to defend our positions on reliability versus cost savings." Caio Cristo, Director of Infrastructure at Luma Health

Building a best-in-class Kubernetes optimization stack on AWS

With PerfectScale handling workload-level optimization, Luma Health now runs an autoscaling and optimization pipeline across the full Kubernetes stack on AWS. Karpenter manages node-level scaling, KEDA drives horizontal pod autoscaling across RabbitMQ, HTTP, and database triggers, and PerfectScale continuously optimizes the resource requests and limits that determine how efficiently every pod uses the capacity Karpenter provisions.

The combination means Luma Health's Amazon EKS environment scales efficiently at every layer: infrastructure, pod count, and individual workload resources. As the platform grows to serve more health systems and more patients, this stack scales alongside it, compounding the value of automated optimization over time.

Caio and his team continue to work with PerfectScale's support team to refine automation policies across environments, ensuring the right balance of cost efficiency and reliability as workloads evolve. The goal is simple: keep Kubernetes costs predictable while Luma Health pursues its mission of making healthcare work better for every patient.