I build software and then try to find out whether it actually works. "It compiles" and "it's correct" are different claims, and the gap between them is where most bugs live.

What I'm building

• bluehex: a marketplace where Claude and Anthropic practitioners publish verified credentials and clients can find them. I lead it as the commercial arm of Code.Sydney, run in the open with contributors across Australia, South Africa and the US. Next.js, TypeScript, Supabase.
• latchkey: four years in property management taught me that tenancy disputes are arguments about the past: what was owed on a given date, when a repair was reported, what condition was recorded at entry. Current state answers none of them, so latchkey reconstructs a tenancy's full history from an append-only log, tamper-evident enough to stand as evidence at a tribunal hearing. It's also where I'm learning event sourcing and domain-driven design properly.
• snag: an FMEA-style method for mapping how code can fail: the failure modes a line-by-line review tends to wave through. I planned a CLI and a collaborative backend, then pre-registered experiments showed the value was in the method rather than the tooling around it, so I cut it back to a Claude Code skill.
• treetop: a Go CLI for working across several git worktrees at once, tracking branches and open PRs so parallel work doesn't lose context.

How I work

• Pre-registered experiments before I trust a result.
• Decision records, so later I can see why I chose what I chose.
• Honest write-ups, including when the answer is "this didn't work."

Stack

Elixir and Phoenix these days, TypeScript and React from before, PostgreSQL throughout. Go for command-line tools; treetop is where I'm learning it.