Built a full-scale SaaS product by shifting from writing code to writing specs — rigorous PRDs and feature briefs designed specifically for AI agents to execute. The primary language wasn't TypeScript; it was clarity. Because I chose the architecture and dictated the patterns upfront, the result was a production-grade, fully auditable system.

Built HARM (HTTP Archive Replay Machine) to give web extraction clients a practical way to understand their own HTTP traffic without needing me in the room. Clients upload HAR files and can investigate, replay, and stress-test their requests — with insights into cookies, request purposes, and auto-generated code they can actually use.

Discovered ryanturner.com was invisible to AI crawlers — Gemini couldn't extract text while every other site worked fine. Rebuilt the site from a Vite/React SPA to Next.js App Router with static generation, rearchitecting routes and logic from scratch. Deployed a blueprint-driven system with 35+ parallel sub-agents to analyze tasks, map dependencies, and execute the migration.

Built an AI production pipeline for Closet Rebellion — a personal styling service — to turn raw clothing photos into polished lookbook content without a photographer or studio. The pipeline cleans up product shots (removing hangers, shadows, backgrounds), then generates a virtual model with custom body attributes wearing the outfit. At roughly four cents per image, it's a practical alternative to traditional fashion photography that scales.