I build software that holds up on a bad day.

How I work

What you can count on

• I build for the worst conditions, not the demo. JobsiteHQ works in basements with no signal and saves a recording every second, because a dropped connection or a dead battery on site shouldn't cost anyone their work.
• I measure instead of guessing. When I report a number, it's the median of repeated runs, not the best one. My last store went from 82 to 95 on Google's mobile speed test, measured across five runs.
• I check the work, including the AI's. I use AI tools to build faster, and I review everything they produce. I wrote about what they get wrong and how I catch it.
• Your money and records are handled properly. Payments run on Stripe and go straight to your bank, never through me. Data is backed up every night.
• You can reach me. No call center and no ticket queue. Email me and you get the person who can fix it.

What I've built

JobsiteHQ

A punch list and invoicing app for handymen and small-crew contractors. Walk the job, talk through what needs doing, and it writes the punch list, the estimate, and the invoice. Runs on a phone, works offline, and the client pays by card from the link.

jobsitehq.app →

Amber Hour Coffee Co.

A production online store for a specialty coffee roaster, selling into the US and the EU in two currencies across six country storefronts. 95 on Google's mobile speed test (median of five runs) and 100 for accessibility.

View the store → · Read the code →

Writing

What an AI agent gets wrong building a Medusa storefront

I built a production store with an AI agent writing the code and me directing and reviewing it. It made the same nine kinds of mistake every time, and handed me a 95 speed score on an easy page while the page customers actually use sat at 82. This is what I found and how I check for it now.