GKE Migration & AI DevOps Platform for iCardio.ai

The challenge

iCardio.ai operates one of the most data-intensive platforms in medical AI — a proprietary dataset of 200M+ annotated echocardiogram images powering deep learning algorithms for automated transthoracic echo interpretation. Their infrastructure had grown organically on AWS, but fragmented tooling, manual deployment processes, and no unified observability were slowing ML model rollouts and creating compliance risk around sensitive cardiac imaging data. They needed a complete DevOps overhaul: migrate 132 TB of medical imaging data to GCP, replatform container workloads from ECS to GKE, and build the CI/CD and monitoring foundation their AI pipelines demanded.

What we did

We established a GCP-native platform from the ground up, with all infrastructure codified in Terraform for repeatable, auditable provisioning across every environment. The 132 TB imaging dataset was migrated from AWS S3 to Google Cloud Storage with minimal downtime, carefully sequenced alongside the transition of containerised workloads from ECS to GKE. Kubecost was integrated from day one to give the team real-time cost visibility across cluster workloads — critical for a company running continuous ML training jobs. GitHub Actions CI/CD pipelines were built to automate testing, security scanning, and AI model deployment end-to-end. For performance, Redis caching cut database load and response latency, while Sentry APM provided real-time monitoring and alerting. Select services were moved to Cloud Run for serverless auto-scaling, and BigQuery was wired in for model training and validation workflows. Security posture was hardened through Prowler-automated cloud audits, IAM governance reviews, and integrated test coverage metrics — giving the team full visibility into both security gaps and untested code paths.

Results