A FinTech AI-driven fraud detection platform partnered with People10 to modernize its CI/CD pipeline and optimize high-performance compute workloads. The client faced high cloud costs due to static GPU allocation, leaving 70–90% of paid accelerator capacity underutilized during AI/ML iteration cycles. Manual deployment coordination also slowed releases and incident recovery.
People10 implemented a Kubernetes-native, GPU-aware CI/CD architecture, integrating NVIDIA GPU sharing strategies—time-slicing, MPS, and MIG—based on workload needs. Automated release workflows using GitLab, Nexus, Trivy, SonarQube, and Helm were supported by Volcano scheduling and Karpenter for dynamic autoscaling.
The modernization delivered ~$500K in cloud savings within three months, along with 60% faster release velocity, 70% faster incident resolution, and 3x higher GPU utilization for light inference workloads.