A Raft-consensus secrets platform wired straight into the private-cloud build pipeline — and a hard look at where it needed to go next.
The trust layer, made highly available.
Secrets are the trust layer everything else quietly depends on — so the platform holding them can't be a single point of failure, and it can't be an afterthought bolted on beside the build. The team needed a Vault deployment that stayed available, authenticated against corporate identity, and fed the private-cloud pipeline directly.
I designed and built the team's HashiCorp Vault HA infrastructure: a Raft-consensus cluster behind a load balancer, integrated with Microsoft Entra ID for identity, and wired directly into the VCF build pipeline so provisioning could pull secrets as part of the automated flow.
Raft gave the cluster integrated consensus storage without an external dependency; the load balancer kept it reachable through node failures; and Entra ID meant access followed the same identity the rest of the org already trusted.
The interesting part wasn't standing it up — it was recognizing where the running design was quietly wrong.
As platform SME, I authored the proposed target-state architecture — an internal RFC — for the secrets platform: Aurora PostgreSQL as system of record, Vault running in ECS containers, lifecycle-managed via GitHub Actions and Terraform. The RFC diagnosed an architectural mismatch in the running design and defined the path to correct it. Proposed architecture — documented and put forward, not yet shipped.