Showcases

Aura X Capital runs liquidity vaults for tokenized real-world-asset funds from Centrifuge — deJAAA, deJTRSY and deHYB, with subscription and redemption in USDC. As CTO I own the architecture: how the on-chain side meets the backend, how it holds up as load grows, FinOps, security boundaries, and the engineering process across roughly 20 services.

Underneath it, an empty AWS account turned into a platform: EKS and FluxCD on Terraform, Prometheus, Grafana and Loki for visibility, one set of reusable GitHub Actions workflows shared by those services, and CDNN — a custom CDN I designed so the product stays reachable where third-party CDNs are blocked.

Reusable GitHub Actions workflows with the Argo family as the core, so a new service onboards itself instead of waiting on an infrastructure ticket.

CI time down 9%, config drift gone, and the same pattern now runs across 6+ client environments.

Observability stacks and Golden Signals dashboards built across four enterprises, so an incident starts with a question that has an answer.

On the same platforms, node warm-up for 1000+ pods went from three hours to seconds.

The same platform pattern at both ends of the scale: a DeFi startup where I set up everything from the first EKS cluster, and a 200-person FinTech running 100+ microservices.

Self-service CI/CD, GitOps and observability from day one, so infrastructure grows with the product instead of blocking it.

Scope-of-Work based assessments and MORA/MODA analysis for multiple clients, ending in both a target architecture and the migration architecture that gets you there.

The deliverable is a path from the current state to production-ready cloud, not a slide deck.

Autoscaling and bin packing with KEDA and Karpenter, preemptible capacity for dev and staging (−10% compute), plus cost visibility and right-sizing.

Infrastructure costs down about 5% where it mattered, with reliability untouched.

Plan and apply move into the pull request: reviewers read the actual terraform plan output, approval gates the apply, and state stays consistent.

Infrastructure changes become reviewable instead of “run plan locally and hope”.

A company-wide Terraform registry of versioned modules, plus IaC training for developers and QA, so teams consume infrastructure as code without a middleman.

Same idea applied to access: Cloudflare Zero Trust defined in code and shipped through CI/CD, so boundaries are explicit and auditable.

Inference on Kubernetes with GPU and spot capacity, right-sized and measured: scaling behaviour and pipeline metrics are visible rather than assumed.

The point is that cost and performance stop being a black box, not that they drop to zero.