Blog

Notes on building

Infrastructure, AI, tools, and the decisions behind them.

Architecture diagram. Three sources on the left — a server session, a Windows box and an occasional third machine, each with its own context window — connect by dashed lines to a central node labelled THE STORE, holding more than 600 notes on a network drive. From there a line leads to a panel headed "Before every request", explaining that the lookup runs per request rather than once per session, extracts the load-bearing terms and shows the five best hits. Badges note a measured runtime of 0.7 seconds, read-only access, and quiet failure.

A Memory That Maintains Itself

I work with several machines and several AI sessions at once. One on the server, one on the Windows box, occasionally a third somewhere in between.…

Four task cards — a multi-stage price calculation, four talks into four slots, six fields into JSON, and a five-seat constraint puzzle — above a full-width bar filled completely and labelled 48 of 48. Result badges note that all 48 cells were correct, that the winning model was nine times faster and half the size, and a panel explains that a model solving everything tells you only that it did not fail.

48 out of 48 — And Why That Is a Problem

Until today, every measurement in this series ran on a single graphics card. 24 gigabytes, and the boundary was clear: anything larger simply went unmeasured. As…