02 / PROJECTS

BoostT1D

Founder and developer · 2024 to present

2,000+ downloads

React
Next.js
TypeScript
PostgreSQL
Prisma
Swift
Node.js
Vercel

01 / THE PROBLEM

The problem

I was at dinner with three other teenagers who have Type 1 Diabetes. Between us we had decades of experience, and none of us could confidently estimate the insulin dose for the food in front of us. Existing apps could record data, but they rarely helped answer the questions that matter most: why did this happen, what can I learn, and what should I do differently next time? I went home and decided to build something that could.

02 / WHAT I BUILT

What I built

BoostT1D is an iPhone app and a Next.js site. The app connects to the CGM setup you already have: Nightscout, Dexcom Share, or LibreLinkUp. If you run an automated loop such as Trio, Loop, or AndroidAPS, it understands which insulin the algorithm delivered and doesn't treat that as decisions you made. You can photograph your plate and get a carb estimate with the reasoning behind it, plus protein, fat, and fibre, then adjust anything that looks off before you save it. The app looks across your last few days for patterns worth noticing, like overnight lows or a time of day that consistently runs high, and turns weeks of data into a PDF report for a doctor visit. It also tracks context like stress and exercise, not only carbs. That came from my own data: exams and dance competitions move my glucose in ways carbs don't explain. The site does two jobs. It's the public side: resources for newly diagnosed families, a Buddy Network that matches people with peers, and volunteering. It's also the backend for the iOS app. Every installed copy calls two API routes on that server, one for food analysis and one for insights, so those routes have to keep working no matter what else changes. The day-to-day tools started on the web and later moved into the app. It has 2,000+ downloads.

03 / THE HARDEST PART

The hardest part

Security. This is people's health data, and I had to teach myself how to protect it from documentation, failures, and GitHub: secrets handling, threat modeling, and database permissions. The AI keys live only on the server, and the app never holds one. The AI routes are rate limited, with a daily cap per user. When an install registers with the site it sends demographics only, never a token or password. The admin console sits behind a signed, HttpOnly cookie with login throttling. Shared live glucose views are read-only and expire.

04 / WHAT I LEARNED

What I learned

An estimate is a starting point, not the last word, so the app shows its reasoning and lets you correct it. Dose suggestions sit behind a server-side flag, so I can turn them off without shipping an app update. Database migrations run as part of every build, so a deploy can't get ahead of the schema. BoostT1D is not a medical device, and it says so plainly. In July 2026 I was accepted to Y Combinator Startup School.