Writing.

Notes on building systems, understanding failures, and making software that works.

Start here

  1. Eleven departments, one afternoon: how UCAM Cloud ran UIU's Fall 2026 section selectionUIU's Fall 2026 section selection on UCAM Cloud: eleven departments in one window for the first time, 7 million requests with 8 server errors, a 114 ms p99, 29,047 course seats placed and none over capacity.Distributed systems ·
  2. Advising day: the afternoon a whole semester picks its seats at onceWanna know something crazy? A platform that sleeps most of the year, then takes thousands of requests a second in one afternoon and what bot defense did to the flood.Distributed systems ·
  3. Distributed rate limiting at the edge with Cloudflare Durable ObjectsWhy the per-colo rate-limit binding leaked my API limits, and how one single-threaded Durable Object per key running a sliding window counter fixed it.Infrastructure ·
  4. Postmortem: DocPipe v1 hung under loadA sanitised postmortem for the DocPipe v1 service hang — root cause, timeline, fix, and the soak test that prevents the failure mode from coming back. Google SRE Chapter 15 format.Infrastructure ·

Case studies

  1. UCAM Cloud: course registration that has to be right in the first minuteHow I built the first production version of UIU's course-registration platform in three months: registration that took five to six hours now takes most students under a minute.Case study · Oct 2024–present
  2. Atlas: the cloud behind source control for coding agentsHow I designed and built Atlas's cloud backend on Cloudflare: sync that accepts the same work twice, Shared Threads for people and agents on one task, and per-organisation isolation.Case study · Jul 2026–present

All writing

esc

Recent