Skip to content

← All projects

2022 → 2024 SRE & DevOps

Authlete — OAuth/OIDC SRE for Japanese banks

I operated and industrialised Authlete's deployments across three delivery models — on-prem, isolated SaaS, shared SaaS — for Japanese banking clients.

  • GCP
  • Kubernetes
  • Helm
  • Terraform
  • GitLab-CI
  • Java
  • JUnit
  • Ruby
  • OAuth 2.0
  • OpenID Connect

Context

Authlete sells OAuth 2.0 and OpenID Connect authentication to banks, primarily Japanese. Three delivery models coexist: on-premise (the client hosts), isolated SaaS (a dedicated instance per client) and shared SaaS. Each model has its own security, networking, and compliance constraints.

The challenge: industrialise these three models with a single template foundation, keep legacy versions supportable for as long as clients need them, and hit the OAuth / OIDC compliance bar on every release.

My role

SRE & DevOps. I owned the deployment templates, the multi-version CI/CD pipelines, and the SaaS operations on GCP.

What I built

  • SaaS deployments on GCP operated daily
  • Kubernetes templates with Helm + Terraform, designed to be cloud-agnostic so they could also serve on-prem clients without major code branching
  • GitLab-CI pipelines capable of building and testing every legacy version of the service in parallel — because you can’t ask a banking client to migrate overnight
  • Java JUnit test suite: unit + integration
  • Ruby test suite: end-to-end + performance
  • OAuth 2.0 / OpenID Connect compliance tests integrated into the pipeline, because an auth service that drifts from the spec stops being an auth service
  • A massive cut in the time between “a dev pushes a fix” and “the fix runs in SaaS at the client”

The hard part: managing the matrix (delivery model × supported version × target cloud) without turning it into unmanageable configuration debt.

Results

Dev productivity sharply increased: changes flow through the automated test chain instead of waiting for manual QA cycles. The same template base serves SaaS and on-prem, which lowers the cognitive load on the SRE team.