Case Study

Cloudchipr Client

How a vector database platform cut cloud spend by 40% with Kubernetes, GitOps, and automated FinOps

Challenge

The client, a vector database and ML data management platform, was running production workloads on legacy GCP infrastructure with heavy manual operations. Releases depended on manual steps and coordination, deployments were prone to failures and rollbacks, and configuration drift accumulated over time without Infrastructure as Code. Costs grew unpredictably without reliable attribution or actionable visibility.

At the same time, the team needed AWS to take advantage of cloud credits without disrupting production workloads or customer pipelines.

Solution

The client partnered with Cloudchipr to modernize the platform end-to-end, migrating workloads from GCP to AWS and establishing a cloud native operating model built on Kubernetes and GitOps.

The solution centered on:

  • Migrating to AWS EKS (Kubernetes) as the orchestration foundation with multi-AZ resiliency and workload isolation
  • Implementing Infrastructure as Code with Terraform for declarative, repeatable infrastructure management
  • Enabling GitOps-based deployments with Argo CD and packaging with Helm for consistent, auditable releases across environments
  • Adding intelligent, event-driven autoscaling with KEDA (including scale-to-zero) to eliminate idle capacity
  • Establishing unified observability across metrics, logs, and traces with Prometheus, Alertmanager, Grafana, Loki, Tempo, and OpenTelemetry
  • Modernizing ingress and traffic management by migrating to Envoy (Gateway API) and automating TLS with cert-manager
  • Introducing fine-grained authorization with OpenFGA and strengthening cloud security posture with AWS GuardDuty
  • Deploying Cloudchipr’s FinOps platform to deliver real-time cost visibility, governance, and automated optimization workflows

Impact

By combining an AWS migration with Kubernetes standardization, GitOps automation, and FinOps governance, the client reduced monthly cloud spend by 40% while improving delivery reliability and operational confidence.

Engineering teams shifted time away from repetitive operational toil and toward product delivery, supported by automated deployments, drift detection, scale-to-demand workloads, and unified telemetry.

By the numbers

  • 40% cloud cost reduction (from $20k to $12k/month)
  • Zero-touch deployments with GitOps replacing manual release processes
  • Unified observability across metrics, logs, and traces with OpenTelemetry integration
Challenges:
Cost, Observability, Scaling
Industry:
AI
Cloud Type:
Public
Published:
July 21, 2026

Projects used

Prometheus logo OpenTelemetry logo OpenFGA logo Kubernetes logo KEDA logo Helm logo Envoy logo cert-manager logo Argo logo

How a vector database platform eliminated manual operations and cut costs by 40%

The organization operated their infrastructure with manual processes on Google Cloud Platform (GCP). As the platform scaled to support enterprise AI/ML workloads, the team encountered critical operational challenges: every release required manual intervention and coordination, deployments frequently encountered issues requiring rollbacks, infrastructure changes were error-prone and time-consuming, and cloud costs grew unpredictably without clear attribution or optimization opportunities.

The engineering team spent significant time on operational toil rather than product development. Without Infrastructure as Code, configuration drift was common. The absence of automated deployments meant releases were stressful, risky events requiring careful coordination and manual verification steps.

To compound these challenges, the team needed to migrate their entire infrastructure from GCP to AWS to take advantage of cloud credits, all while maintaining service availability and avoiding disruption to customers running production AI/ML pipelines.

“Before, every release depended on manual steps and coordination, which made deployments risky and inconsistent. Moving to GitOps on Kubernetes gave us repeatable, auditable releases, and the cost savings finally became measurable and actionable.”

Vlad, Lead DevOps engineer, Vector database platform

Building cloud native resilience with Kubernetes and GitOps

The team orchestrated a comprehensive infrastructure modernization, migrating workloads from GCP to AWS while implementing cloud-native best practices. The new architecture leverages Kubernetes on AWS EKS as the foundation, with declarative GitOps workflows ensuring consistent, automated deployments. The team built comprehensive observability across all layers, while the Cloudchipr FinOps platform delivered unprecedented cost visibility and automated optimization.

Platform and foundation

Why Kubernetes and KEDA: orchestration + scale-to-demand

Kubernetes provides the core orchestration layer with declarative manifests, self-healing capabilities, rolling updates, and workload isolation. The platform uses namespaces and RBAC policies to separate environments and control access.

KEDA enables event-driven autoscaling based on custom metrics rather than just CPU and memory. The team implemented MongoDB-based scaling, where KEDA monitors record counts and processing queues to scale workloads up during heavy data processing periods and down to zero during idle times. This eliminated the need to over-provision resources for peak capacity.

Envoy serves as the API Gateway after the team migrated from the deprecated Ingress NGINX controller. Envoy provides advanced traffic management, observability, and security features, including rate limiting, circuit breaking, and automatic retries. The migration to the Gateway API future-proofs the platform’s ingress architecture.

Why Argo CD and Helm: GitOps for consistent deployments

Argo CD implements GitOps principles, treating Git repositories as the single source of truth for infrastructure and application definitions. Argo CD continuously monitors Git repositories and automatically syncs changes to the cluster, eliminating manual deployment steps. The platform detects configuration drift and automatically remediates it, ensuring the actual cluster state matches the desired state defined in Git.

Helm packages applications as reusable, versioned charts with environment-specific value files. The team maintains a library of Helm charts for common application patterns, enabling consistent deployments across development, staging, and production environments. Helm’s rollback capabilities provide safety nets for failed deployments.

Cert-manager automates TLS certificate provisioning and renewal from Let’s Encrypt and other certificate authorities, eliminating manual certificate management and preventing expiration-related outages.

Why Prometheus, and OpenTelemetry: observability for performance and cost

OpenTelemetry provides the unified collection layer for metrics, logs, and distributed traces. By standardizing on OpenTelemetry, the team can instrument applications once and export telemetry data to multiple backends without vendor lock-in.

Prometheus scrapes and stores metrics from Kubernetes infrastructure, applications, and the control plane. Custom ServiceMonitor definitions enable automatic discovery of new services. Prometheus writes long-term data to S3 for retention and historical analysis.

Grafana consolidates visualization across all telemetry signals. Dashboards provide visibility into cluster health, application performance, resource utilization, and business metrics. Grafana’s alerting rules notify the team of anomalies before they impact customers.

Loki aggregates logs from all cluster workloads and infrastructure components. Logs are stored in S3 for long-term retention.

Tempo captures distributed traces, allowing the team to track requests across microservice boundaries – critical for debugging performance issues in complex, distributed AI/ML data pipelines.

Alertmanager handles alert routing, grouping, and notification delivery from Prometheus. It deduplicates and groups related alerts, routes them to appropriate channels (Slack, PagerDuty, email), and manages alert silencing during maintenance windows, ensuring the team receives actionable notifications without alert fatigue.

Automated cost hygiene and governance

The SaaS platform delivered comprehensive visibility into AWS costs and usage patterns. The team implemented consistent resource tagging policies across all infrastructure, enabling accurate cost attribution to specific teams, projects, and environments.

Automated cost optimization workflows became a cornerstone of the solution:

The platform provides real-time dashboards showing cost trends, budget forecasts, and anomaly detection. Alert policies notify the team when spending exceeds thresholds or anomalies are detected, preventing budget surprises.

Results