Process
Six principles behind every project on this site — how a rough idea becomes something real users run in production.
01 / 06
Before a line ships, I decide the shape — a data model, an ADR for every load-bearing choice, and a single source-of-truth doc the whole build answers to. The decisions get written down first, so another person could pick up the spec and build from it. The repo reads like a design doc that happens to compile.
I build inside the Claude Code CLI, wired to MCP servers that drive the database, the browser, and n8n as first-class tools. Automations ship as version-controlled code, pushed through an MCP — never clicked together in a UI.
Speed is fenced in on purpose: PreToolUse hooks block stray writes, custom skills enforce house conventions, and ADRs keep decisions from drifting. The agent moves fast — inside walls I built first.
The model lives only where it earns its place. The load-bearing core — money in integer cents, server-side RBAC, append-only audit logs written in the same transaction as the change — stays deterministic, typed, and provable.
Each push runs the gate: unit, integration, and end-to-end tests, plus Snyk scanning and automated dependency patching — all in GitHub Actions. It's green, or it doesn't ship.
Production on Vercel, Supabase, Neon, Render and Stripe, with Prometheus and Grafana watching it run and CI/CD deploying continuously. Real users, real money, watched continuously.
Repos are private by design — client IP and live products. Happy to screen-share the codebase, ADRs, and CI on a call.