How it's built

Several apps sharing one account and one library

Picashot is several apps sharing one account and one library, which is most of the reason it took as long as it did.

Mobile app: iOS and Android, one Expo codebase
picashot.co: the main site
Studio: creator tools, separate app
Portfolios: rendered on creators' own domains
Origin: reverse image search for credit
Blog: editorial and release notes
API: one backend behind all of it
Workers: thumbnails, hashes, embeddings

It started as an Android app in Kotlin, because that's what I knew. A year later I rewrote it in Expo so it could run on iOS too. That meant one codebase for both platforms and over-the-air updates, so a fix can reach people the same day instead of waiting on a store review.

The website has been rewritten twice. Next.js, then Next.js again, and it now runs on TanStack Start. The site is mostly image loading and data fetching, and Start handles both without getting in the way.

The backend is a Node API split by feature with Postgres behind it. Images are the expensive part. An original goes to S3, a Lambda picks it up and generates five sizes with Sharp, and a webhook writes the URLs back when it's finished. Uploads don't wait for any of that. Perceptual hashes and CLIP embeddings are generated on the same path.

Expo, React Native: the app, updated over the air
TanStack Start: the website
Node, TypeScript: the API, split by feature
Postgres: photos, games, people, points
pgvector: image embeddings for Origin
S3, Lambda, Sharp: originals in, five sizes out
SES: login codes, digests, Wrapped