Internal Developer Platform @ project44

The problem: infrastructure as a bottleneck

When every team needs the platform team to hand-provision infrastructure and wire up deploys, the platform team becomes the bottleneck and every team reinvents the same wheel slightly differently. The fix isn't more tickets — it's paved roads: standardized, reusable building blocks teams can self-serve.

The golden path

I built the reusable layer that sits between raw AWS/Kubernetes and the product teams:

• Terraform modules — parameterized infrastructure primitives teams compose instead of writing from scratch.
• Helm charts — standardized service packaging with sane, safe defaults.
• Argo CD ApplicationSets — GitOps delivery that templates deployments across clusters and environments.
• GitHub Actions workflows — reusable CI so every team's pipeline follows the same guardrails.

Observability standards

Reliability doesn't come from hope — it comes from seeing problems before users do. I built the Prometheus and Grafana layer as dashboards and alerts as code, so every new service inherits a real observability baseline instead of depending on whoever remembers to build one.

• SLO dashboards make error budgets visible, so teams know when they can ship fast and when to slow down.
• Alert tuning keeps signal-to-noise high — a page should mean something and have an action behind it.
• Dashboards and alert rules live in version control, reviewed like any other production change.

Platform networking: ingress-nginx to Gateway API

Ingress ran out of room. Every real feature — rewrites, timeouts, canaries — lived in vendor annotations, and routing config shared one object across platform and application concerns. Migrating to Kubernetes Gateway API backed by AWS Load Balancer Controller fixed the ownership split.

• The platform team owns Gateway resources — listeners, certificates, and who may bind to them.
• Application teams own their own HTTPRoute resources, without touching shared traffic config.
• Standardized how new services connect to the cluster, ahead of the industry-wide shift off Ingress.

AI-assisted operations

I work AI-first, on a strict plan then generate then review-every-diff discipline. Reusable agent workflows built with Claude Code, Codex, and MCP integrations take the toil; the judgment stays human.

• Auto-generated runbooks from incident timelines, reviewed and adopted as the team template.
• Alert summarization and assisted incident triage during live incidents.
• Custom agent skills and MCP integrations for repetitive Terraform, Kubernetes, and CI/CD work.