Loading articles…
Notes on systems, AI and the code in between
Most software writing stops at the moment things work. I have always been more pulled by what happens next: the 3 a.m. surprise, the assumption that quietly stopped being true, the fix that took one line and two days to find.
Signature is my recollection of those moments. Every article begins with a problem I actually sat in front of, and I try to put you in that chair with me. You get the first hunch, the theory I was sure about, the thing that proved me wrong, and the quiet click when it finally made sense.
I leave the dead ends in on purpose. They are usually where I learned the most, and a tidy write-up that hides them makes the answer look easier than it was. If I lost an afternoon to a wrong turn, you should get to skip it, and see why it was tempting.
I write because explaining something is how I find out whether I really understand it. If a piece leaves you with one idea you can use on Monday, or just the feeling that someone else got stuck here too, it did its job.
Three recurring threads, each written to be read in one sitting:
Caches that stampede, queues that back up, migrations that go quiet. What breaks under load, and why.
What changes when a model becomes one component among many, with timeouts, retries and a budget.
Short, practical write-ups for the problem you are staring at right now.
No sponsors, no funnels, no listicles. Articles are published when they are ready, corrected in the open when they are wrong, and written for a reader who is smart but short on time.