Search ads · AI · Launched
AI-designed search modules
An LLM designs new ad modules for eBay search, and small, fast models run them live. Two are in production worldwide.
+$194M annualized GMB · +$4.9M ads revenue
Read the case study →Case studies
Launched products and working prototypes. Each one covers the problem, the bet, how it was built and what happened.
Launched at eBay
Search ads · Launched
eBay’s first ad modules in search, and the four that followed
Sponsored carousels on eBay search, each built around one reason to buy. Four launched, adding about $18M a year in ad revenue; a fifth is confirmed for launch.
≈ $18M a year across four launched modules
Read the case study →
Search ads · Video · Launched
Video in the most visible ad slot on search
The Featured Item Card gives one sponsored item two to four times the space of a normal result. I added video to it, designed the quality controls and ran the tests. Live worldwide, adding $3.6M a year.
+$3.6M a year in ads revenue
Read the case study →Prototypes
Search · AI · Prototype
The best listing for every search, explained by AI
A custom reasoning engine reviews the top results for a search, picks the one listing a buyer can trust fastest, and makes the case for it.
Read the case study →
Search · Marketplace · Strategy + prototype
Showing shoppers what theirs is worth
While people browse, eBay already knows what those items sell for. These modules point it out: “these go for about $50. Have one?”
Read the case study →
Seller growth · Ads · Prototype
Bounties for what buyers want most
List an item eBay knows is in high demand, and earn ad credits to boost everything else you sell.
Read the case study →
Personalization · AI · Prototype
A homepage that plans itself
Three ways to build a shopper’s homepage, side by side: fixed, rule-based, and planned by an LLM, with a “why this?” on every module.
Read the case study →
Discovery · AI · Prototype
Interesting eBay
A feed of the weird, rare and suddenly popular things on eBay, for browsing rather than searching.
Read the case study →Side projects
Built on my own time with AI coding agents, on real production stacks. How that works is on How I build with AI.
My business · Client portal and website · Side project
Safe As Houses: a client portal and website for my digital security advisory
Safe As Houses helps people make their digital lives secure and recoverable. I built its client portal and public website alone with Claude Code. Since late September, every change has gone through the same review-and-test pipeline a software company would use.
354 commits since March · 91 pull requests since late September
- Next.js
- Supabase
- PostgreSQL
- Anthropic API
- Vercel
Fitness · Phone app · Side project
Aether: a strength-training app with no App Store and no subscription
Workout logging and well-known training programs in one free app. It installs from Safari, works offline and keeps its data on the phone.
12 training programs · 258 exercises · $0 to run
- React
- TypeScript
- Progressive web app
- Vitest
This website · Next.js on Vercel · Side project
This website, built in conversation with Claude Code
A portfolio built on my business’s design system: pre-rendered, fast, and ready for live demos.
Live two days after the first line of code
- Next.js
- React
- Tailwind CSS
- Vercel
Music · Spotify · Prototype
Monolith Ultra: an AI planner for a messy Spotify library
It reads a large Spotify library in stages, lets me set firm limits first, then has Claude plan new combined playlists, without ever deleting an original.
- Python
- FastAPI
- React
- Anthropic API