Oracle · 2016–2025 · Founding architect

Oracle Verrazzano

Verrazzano is Oracle's open-source enterprise platform for running applications across many Kubernetes clusters. It reached general availability in 2021. Ravi Singhal was its founding architect.

The problem

Enterprises ended up with Kubernetes clusters in several places at once: on-premises, in Oracle Cloud, and in other clouds. Each cluster needed the same supporting stack — service mesh, metrics, dashboards, tracing, logging, search, identity — and each team installed and upgraded that stack by hand, differently. The operational cost was not in running one cluster. It was in keeping thirty of them consistent.

What he built

He designed the multi-cluster architecture and built the control plane in Go, driven by Kubernetes operators. The platform installs and reconciles a nine-component stack through declarative custom resources: Istio, Prometheus, Grafana, Jaeger, Kiali, OpenSearch, Rancher, Keycloak and Fluent Bit. An application and its supporting infrastructure are described once, as data, and the operators converge every registered cluster onto that description.

The design decision that mattered was making the operator the only writer of the state it owns, and making everything else a declarative input. Reconciliation loops that fight each other are the common failure mode in platform operators, and most of the architecture work was deciding which controller owns which field.

Adoption and outcome

Verrazzano grew to more than 200 tenants, ten internal Oracle teams and five external enterprises. Ravi closed its first enterprise deal, worth $1M with a large Japanese enterprise, working the sales cycle forward-deployed as the solution architect and handling gap analysis and compliance engineering. It was the platform's first revenue outside open source, and it shaped the product roadmap afterwards.

He built and led the twenty-engineer platform team behind the product, owning hiring and technical direction.

The operator patterns here carried forward directly intollm-inference-platform, which applies the same single-writer discipline to GPU inference workloads. More on Ravi's background is on theabout page.