Site Reliability Engineer
American Express
Jun 2022 – Jul 2024 · Atlanta, GA
CONTEXT
Primary role: SRE owning shared AWS infrastructure across 6 PCI-supporting production accounts - EKS, RDS/Aurora, Lambda, IAM, and multi-account VPC networking - with no systematic cost controls and cloud spend climbing. On the side, developers constantly hit the team for manual infrastructure operations (SQS dumps, vault secrets, on-call contacts), creating a steady stream of interrupts.
THE WORK
WHAT I TOOK ON
Own the AWS platform at 99.9999% uptime under 24×7 on-call in a regulated PCI environment, reclaim $200K+/month in cloud waste, and automate away the manual developer requests - all simultaneously.
WHAT I DELIVERED
AWS Platform: Terraform-managed infrastructure with Python and Bash automation for rightsizing (oversized instances, idle resources, off-hours autoscaling) cut $200K+/month. 24×7 on-call across Datadog, Dynatrace, and Splunk held <0.01% error-rate SLOs; CI/CD pipelines and quarterly DR exercises kept the platform resilient. Also built DevDash: a TypeScript/Node.js approval-gated platform where Lambda handles SQS dumps, vault secret delivery, and PagerDuty paging - eliminating the manual developer request queue entirely.
KEY CONTRIBUTIONS
- Operated shared AWS platform across 6 PCI-supporting production accounts at 99.9999% uptime: EKS, RDS/Aurora, IAM, S3, Lambda, Transit Gateway, multi-account VPC networking
- Built Python and Bash automation for instance rightsizing, idle-resource cleanup, and off-hours autoscaling - cut cloud spend by $200K+/month
- Led 24×7 on-call using Datadog, Dynatrace, and Splunk for incident detection and triage; held <0.01% error-rate SLO; cut MTTD 50% and MTTR 35%
- Managed infrastructure-as-code with Terraform and maintained CI/CD pipelines for platform services
- Planned and ran quarterly DR exercises, failing over Aurora/RDS and EKS within 8-hour RTO/RPO targets
- DevDash: SQS → S3 workflow, vault secret delivery, and PagerDuty paging via approval-gated Lambda - built in TypeScript/Node.js, used by 50–100 engineers
EQUIPMENT MANIFEST