Writing and press
Things in print, and a few things I wrote.
Duke Pratt School of Engineering · Oct 2026
AI-Driven Product for Builders
Coverage of EGR 590, Duke’s summer AI product sprint, including the tolerance and design-review tool I built after my surgical robotics internship. The tool is ToleranceForge.
Read on pratt.duke.eduSSRN · 2023
Ride Like the Wind Without Getting Winded
Quantitative modeling on the growth and effects of electric bikes in the 21st century. Co-authored. Ranked in the top 10 downloads in three SSRN categories at the time of posting.
Read on SSRNNotes from the field
Short takeaways from the lab, the shop floor, and the deal pipeline.
Not a blog. A running list of things that turned out to be true more than once.
Lab
A design is not done until it can be ordered.
At DEVOL, the moment a circuit became real was when the parts list went to purchasing. Everything before that is a sketch, and the sketch is where the creative work lives.
Industry
Tolerances are a budget line.
At Asensus, a tolerance nobody questioned cost more than any part on the drawing. Nobody owned it because it did not look like money.
Investing
Ask what breaks at 10x.
Most pitches describe the system at its current size. The diligence question is what fails when volume, headcount, or regulation scales by an order of magnitude.
Teams
Build what outlasts you.
Four years on an FRC team with no playbook taught me that onboarding is the product. The robot is secondary.
Clinic
The clinician is usually right.
A prosthetist solved the leg support problem on a cat wheelchair in one conversation after I had spent weeks on it. Design intuition is a hypothesis, not an answer.
Shipping
Cut scope until it hurts, then ship.
ToleranceForge only became real when I dropped 3D parsing and live supplier APIs. The small version that works beats the big version that does not.