I build cloud infrastructure
that doesn’t wake people up at 3am.

My day job is keeping the developer platform boring — GitHub Enterprise, JFrog Enterprise, and Jenkins Enterprise, the boxes every engineer at the company quietly depends on. Cloud infrastructure is shaped with AWS CloudFormation; the codebase around it is being pulled into SOX and enterprise-compliance posture, which mostly means tightening logging and making audit trails something an auditor can actually read.

The load-reduction work on GitHub Enterprise started as a profiling exercise. Three months of watching traffic, isolating the top consumers, and walking specific teams through targeted optimizations — not a rewrite, a conversation. The 120K requests an hour that stopped hitting the platform did so because the callers changed, not because the platform did.

Around that, a few smaller systems. A telemetry service for the golden signals on GitHub instances, written against Telegraf with Starlark for the shaping logic, so latency, traffic, errors, and saturation become something a dashboard can answer for. StackStorm workflows for the operational edges — the steps an on-call would otherwise run by hand. And an audit tool for Just-In-Time access on GitHub, built so access reviews stop being a manual spreadsheet exercise.

A year on the Azure side of the fence, mostly writing Terraform for clients who needed their infrastructure to be repeatable before it could be anything else. Modular Terraform as the substrate; multi-stage Azure DevOps pipelines on top — environment-specific variables, approval gates, compliance checks wired into the same definition that did the deploy. Kubernetes workloads ran on AKS, packaged with Helm, fronted by Nginx ingress.

The architecture I spent the most time on was an engagement called Avanos — a private data estate stitched across site-to-site links from Azure back into the customer's datacenters, then through Data Factory, Data Lake, and Synapse, ending in PowerBI. Every hop was designed to never leave private space: Private Endpoints and Private DNS Zones carried the traffic. The Hub-and-Spoke work, the APIM-fronted RAG chatbot over Azure AI Search, the WAF and firewall posture — all of it came from the same instinct, which is that security and networking aren't a late-stage concern, they're the shape of the diagram.

First real engineering job, sitting between the application and the database, learning what production actually asks of you. The largest piece was a MudBlazor tool that gave enterprise clients a view into their cloud databases — inventory, health, the usual — plus a predictive layer meant to catch problems before an on-call did.

The rest of the time went to a regulatory-repository and ad-hoc reporting backend for a PwC engagement, built on Azure DB and .NET 8, and some early work wiring ChatGPT's API into internal servers and document systems back when that still felt unfamiliar.