Notes from the PDF factory
Essays and technical write-ups on AI-authored PDF generation, catalog-driven architecture, and the long road that got us here. Every post is also available as a download-ready PDF — rendered by the same engine that powers the public API.
- Jason Galea with Claude Opus
Four goes at the same idea
I've been bumping into PDF generation as a problem since 1999. Four of my attempts to turn it into a product have reached the level of "you could use this." Three didn't find a market. This is the history, honestly told.
historyarchitecturelessons - Jason Galea with Claude Opus
The builder I didn't build
The previous three versions of this product all assumed a template-authoring UI was the hard thing to build. The 2026 version doesn't have one. An AI reads a sample document and produces a template directly.
architectureailessons - Jason Galea with Claude Opus
A catalog, a tiny DSL, and a skill file
Making AI reliably produce a working PDF template needs three pieces working together — a machine-readable catalog, a compact DSL that costs 75–95% fewer tokens than raw JSON, and a skill file the agent can fetch once and remember.
architectureaitechnical - Jason Galea with Claude Opus
Sub-second, no eval, no browser
How a PDF engine fits inside a single Cloudflare Worker request with no eval, no headless browser, and no Node filesystem — and still returns sub-second renders for typical documents.
architectureruntimetechnical