Case Study — P-01
A field guide to AI red teaming
A live resource that turns a sprawling subject into something you can actually learn. Designed, written, and built end to end.
A working method, 95 techniques, and a pile of exercises. The content existed. The problem was the architecture: how do you turn all of it into one thing a person can actually learn, front to back?
Design decisions
One path, not a reference dump
Forty pages of material could have been a wiki. I built it around one method instead, a five-phase loop, so a newcomer follows a path and comes out able to think, not just look things up.
Turn the catalog into a tool
A big catalog of options is a table nobody reads. So I built Compose, a tool. You pick, you stack, and it generates the result for you. The reference turned into something you do.
Design for a long, dense read
People read this site, they don't skim it. I capped the measure near 66 characters, kept a steady rhythm, and gave every page an on-this-page index, so the density never lands as a wall of text.
Performance is part of the UX
Astro ships zero JavaScript by default. React hydrates only for Compose and the hero. Everything else is plain HTML that loads before you notice. On a free resource, speed is the first thing a reader feels.
The catalog was the challenge. Hundreds of options that had to stay teachable without turning into a spreadsheet. Compose went through a few rewrites before it felt simple instead of just complete.
What this shows: I take a messy, sprawling subject and turn it into a fast, clear, usable site. Research, design, content, and build.
I designed and directed the site. The build ran with heavy AI assistance, fitting for a project about testing AI. The method, every decision, and all the content are mine. The typing had help.