Product leader who builds. I run the org and ship the AI myself.
Head of Product at ScanPay (fintech, 0 to 1, employee #3): I built and lead product, design, analytics, and support, and I personally built the AI systems that run them, with Claude Code and Cursor. I'm a product leader who never stopped building. This profile is where I keep the builder side on display: the build-vs-buy calls, how I eval, and where an LLM is dangerous and how I box it in.
| What | The call, and the outcome | Role |
|---|---|---|
| Support-ops AI stack | Ran a support org on a self-built AI stack instead of per-seat vendors, because our SOPs and voice are what a vendor can't sell | Owner, 0→1 |
| Call-QA pipeline | LLM-as-judge scoring of every call against our real SOPs, at cents per call | Built it |
| Mystic Monk rebuild | Rebuilt an agency site from scratch after a malware infection: design, build, and a clean DNS and email migration | Built it |
| UX Proofsheet | A Claude skill that benchmarks a design against how real apps ship, evidence-backed and scored | Built it |
| AI-craft notes | How I think about evals, agent harnesses, and the AI-era PM bar | Author |
- The next layer of the UX Proofsheet: a quantified design-token comparison
- 📓 AI-craft notes · 🔍 System-design write-ups · 📐 Eval rubrics