Building in public, starting now

For eight years I made enterprise systems hard to break — data classification, leak detection, and compliance-bound services at Atlassian Guard and ServiceNow. I'm now pointing that same instinct at frontier AI: red-teaming models, building content-safety classifiers, and shipping guardrails for LLM agents.

This is where I write it down as I go — what I build, what breaks, and what I learn.

The first real posts land in July: a transformer from scratch, then an LLM red-team harness. See you then.