Deep dives & field notes

In-depth writing on LLM engineering — and the systems behind it.

Long-form notes on building, shipping, and speeding up language-model systems. Measured, first-person, and free of hype.

Subscribe to the letter Read the blog →

The current series

Browse the archive →

Test Series · Part 3

Jul 31, 2026 · 4 min read

Every knob we ship: a stress test of the writing stack

A deliberately over-built essay that puts every rendering feature the blog supports on one page — headings, math, a live diagram, an interactive 3D plot, code chrome, margin notes and figures — so nothing can hide.

Continue the series →

Recent writing

Browse the archive →

Batching is a scheduling problem

Once retrieval works, the next wall is latency — and most of the latency you can actually control turns out to be a question of scheduling, not hardware.

ServingSystems

Sep 18, 20251 min

The index is not the memory

An index tells you where things are. It does not tell you what they mean — and confusing the two is where retrieval quietly starts to lie.

RetrievalSystems

May 3, 20251 min

Why my first vector store was a CSV

The cheapest retrieval system that could possibly work, and what it taught me before I reached for anything heavier.