I build and operate a four-component real-time platform — a Discord bot, a Next.js web app, a zero-dependency log-parsing agent, and an auto-updating Electron desktop client — as one monorepo, solo, for a live user base that depends on it during scheduled four-hour events three nights a week.
The system, its four components and four release cadences, the scale it runs at, and four problems worked end to end — a 32-second page, a join that silently matched 2.6% of the time, a pattern compiler that ate its own input, and a plausible answer that turned out to be fiction.
A four-component real-time platform and a party word game — both with real users, both live. What each one does, the measured scale behind it, and links so you can go and look rather than taking my word for it.
The fastest way to reach me, and pointers to the two things people usually want to know before writing.
Every figure across this site is drawn from the git repository, the production database, or the assistant's own usage records. Each page says which, and the numbers are floors rather than best cases — where two measurement windows could be conflated to make something look better, they are listed separately instead.
Spending less by re-deriving less: memory kept as files rather than conversations, retrieval split across two tiers, rules written after paying for their absence, and eight audits that each found a real defect.
The spend, measured rather than argued: metered-equivalent consumption at list rates, a 5.9× gap against the same work run without caching discipline, and the objections answered in the order they usually arrive.