Akul Gopal
Devali

I built a Kubernetes-based distributed replay platform from scratch — owning everything from cluster topology and network architecture to deployment automation. One of the more interesting problems was enabling 1:1 user-to-pod isolation through carefully crafted Services and Ingress routing, so every concurrent session got its own sandboxed environment. I automated the full environment provisioning lifecycle with Jenkins, so new replay instances spun up and tore down without any manual intervention.

The platform scaled to 20,000 concurrent replay sessions with dynamic resource orchestration, and held 99.999% availability under sustained simulation workloads across both Akamai and AWS EKS clusters.

My main focus was making 15+ microservices more reliable and cost-efficient. I built an internal reliability platform that automated capacity benchmarking and surfaced performance trends before they turned into production incidents. On the cost side, I cut AWS compute spend by 25% — mostly through instance right-sizing and shifting 60% of workloads to spot instances with smart scheduling policies.

I also built a load testing framework with K6 wired directly into the Jenkins CI/CD pipelines as a performance quality gate — if a build failed load tests in staging, it didn't go to production, full stop. That alone cut manual testing effort by 40% and kept regressions from slipping through. To close the loop, I set up a Grafana and Prometheus observability stack that gave the whole team real-time health visibility across every environment.