On the Chrome Web Store since April 2026. Weekly users roughly tripled from mid-September to mid-October 2026.

• Types text into a Google Doc or Slides deck at a natural pace and finishes at exactly the time the user picks. It plans every keystroke and pause before it starts, then stretches or squeezes the plan to fit. In an audit of thousands of plans, every run finished within 1 millisecond of its target.
• Added accounts (Supabase) and a $20 one-time Pro license sold through Stripe Checkout. The server checks that each payment notice really came from Stripe, and a notice that arrives twice does not create two licenses. If the license server goes down, paying users keep Pro instead of being locked out.
• Tested against the real thing: 214 unit tests run the code that ships, and an automated check types into a real Google Doc and Slides deck every day. It caught Google losing the last few letters after a page reload (“part way” became “party”). HumDocs now retypes them.

On the App Store since October 9, 2026. Built in Swift, with a small server that reads receipts using Google Gemini.

• Keeps track of who owes whom, one person at a time. No accounts, and the records never leave the phone. Scanning a receipt sends the photo to Google Gemini to read the items, tax and tip, but only after the user agrees on a screen that says so.
• Apple rejected the first version for app quality. I installed it fresh, found the problems (welcome pages skipping, a screen nobody could reach), fixed it, appealed and shipped. I held back the iCloud sync I had built: it was untested on two phones, and a money app that syncs wrong is worse than none.
• Built nine test programs that run the app’s real code, including 20,000 random ledger changes and 50,000 random bills. One showed that when people ordered different amounts, tax and tip nearly vanished: $5.10 of tax split three ways paid out $0.03. Fixed before release.

• Four-week summer program. My team built a Naive Bayes classifier in Python that tells human-written text from AI-generated text, trained and tested on a labeled dataset from Kaggle. Earlier projects used k-nearest neighbors.

• “Likes, Comments, and Shares: How Interactive Features Guide Children’s Moral Development and Understandings of Justice.” A self-directed 12-week review, mentored by a Vanderbilt University PhD student in developmental psychology. Unpublished.

GPA 3.86 unweighted · SAT 1590 (800 Math) · National Merit Semifinalist · AP Scholar with Distinction · AP Computer Science A, Computer Science Principles, Microeconomics and Macroeconomics: 5