Learning

The Library

Subjects I am studying in the open. Each shelf is the same shape: a thing that runs, a written recipe behind it, and a way to rep the thing until it survives being asked cold.

Nothing here is a course and nothing is finished. A shelf earns its place by being repeatable — if I cannot practise it, it stays in a notebook instead.

Open Inference Engineering

Inference Engineering

A dense course and practice ladder on why models are slow and what the levers actually are: bytes, honest timing, profiling, GPU execution, KV-cache math, and serving. Every open lesson has timed drills; the hours are in the problems, not the page length.

5 of 8 modules · 4 notebooks · 22 drills

Course

Interview Prep

A live TypeScript bench, the algorithm spine kept cold-solvable by rotation, practical builds that arrive in phases the way real screens do, whiteboard recipes that draw themselves one move at a time, and role guides for loops that are not a standard SWE screen.

31 recipes · 3 boards · 1 guide

Practice

Shape

Runs before it reads

Every shelf starts with something executable — a bench, a board, a runner — and the writing comes second, explaining a thing you can already press. Notes that describe code nobody can run go stale without ever announcing it.

The test is simple: if a page cannot tell me I am wrong, it is not a shelf yet.

Built to be repeated

Reading something once teaches you that it exists. Producing it cold, from a blank file, on the fourth pass, is the part that survives pressure — so each shelf is arranged for reps rather than for coverage.

That is why the catalogue is short and stays short. A small set held to motor memory beats a long one half remembered.