The best infrastructure work is invisible - nobody files a ticket thanking you for the outage that didn't happen.
A few recurring ideas across fintech, infra, and building this site that didn't each earn their own post.
No posts match this filter.
Open to collaboration, roles in fintech, or backend-focused engineering.
A retry with backoff that quietly absorbs a flaky upstream call. A health check that catches a bad deploy before it takes traffic. A backup that actually restores cleanly when someone finally needs it. None of this shows up in a demo, but it's most of what good infrastructure work actually is.
It's hard to get credit for an incident that never happened, which is exactly why this kind of work is easy to underinvest in. The teams that do it well tend to talk about it anyway, if only so the next person understands why the boring parts matter.
Some ideas show up again and again without ever being big enough for their own post: automate the boring step before optimizing the interesting one, keep the deploy path identical between staging and production, and write the migration you'd want to read at 2am during an incident.
This one's tagged broadly on purpose - it's as much a note to my future self as anything else, a place to point back to when one of these ideas resurfaces on a new project.
Three years into building investment management software, the biggest shift hasn't been technical - it's how much slower I've become on purpose. A feature that would ship in a day elsewhere gets an extra review pass here, because the cost of being wrong isn't a bug report, it's someone's numbers being off.
It took a while to stop seeing careful, deliberate pacing as a productivity problem. In this domain it's the whole point - the fast version of a reconciliation job that's occasionally wrong is worthless compared to a slower one you can trust completely.
Most of the time an ORM is the right call - it's faster to write, easier to review, and keeps queries consistent across a codebase. But every so often a report needs a window function, or a migration needs to backfill a few million rows, and reaching for raw SQL is just faster than fighting the abstraction.
The skill isn't picking one tool forever, it's noticing when the abstraction is costing more than it's saving. A hand-written query with a clear comment explaining why is often more maintainable than an ORM incantation nobody wants to touch.
Working on investment management software changes how you think about correctness. A UI bug is annoying; a reconciliation bug is a different category of problem entirely.
The most valuable infrastructure work I've done is rarely clever. It's automation that removes a manual step, a pipeline that fails loudly instead of silently, and services that integrate front-, middle-, and back-office workflows without anyone noticing the seams.
This site used to be a React SPA that fetched blog posts from a separate CMS instance over REST, with everything else hardcoded. It worked, but it meant two deployments, a CORS workaround, and a growing pile of content that only lived in source code.
The rebuild moves everything into Astro, with EmDash running as part of the same app instead of a separate origin. Content edits now show up without a rebuild, and the whole site - hero copy, experience, projects, contacts, and posts like this one - is editable from one admin panel.