I keep websites up.
Mostly that means cloud and edge infrastructure for a set of high-traffic publishing properties — AWS underneath, Cloudflare in front, Terraform holding it in place. The work is judged by how little happens.
I like the parts nobody writes blog posts about. IAM policies scoped to what a job actually does. Backups that have been restored at least once. Migrations planned early enough that nobody notices them.
Nobody could say what the AWS bill was actually for. Rolled Cloud Cost Management across every account via Organizations and StackSets, pulled in hosting and database spend, and tagged it back to the properties that caused it. The argument stopped because the number was on a screen.
A national platform gets one enormous day a year. Ported the database config to a new engine, restructured five PHP-FPM pools from on-demand to static, retuned nginx per vhost, and sat on the connection graphs through the window. One incident: a query with a leading wildcard exhausted the pool. Found and killed in minutes.
A shared automation role had full admin because scoping it properly was somebody's someday task. Read what the jobs actually did — database dumps, storage sync — wrote policies to match, and swapped them in across three projects. Zero pipeline failures.
Cached pages were going stale in a way that made no sense from the dashboard. The Worker was passing 304s straight through to clients that had never sent a conditional request. Small bug, wide blast radius, satisfying to find.
Seat and usage data from several AI vendor admin APIs, pulled on a schedule into a single Notion database. Python on Lambda, deployed with Terraform, triggered by EventBridge. It runs and nobody thinks about it, which is the point.
Hand-written HTML and CSS in one file. No framework, no build step, no webfonts — the type is whatever your machine already has. It's served from Cloudflare's edge and the numbers at the bottom are measured in your browser, not claimed by me.