Kanta

Basically Google Drive for photos at an event. Everyone shoots on their own phone, the photos land in one album, and face detection helps each guest find the shots they're in. The backend is FastAPI on Azure, with face clustering on AWS Lambda.

Why I built it

After an event, the photos are spread across a dozen camera rolls and group chats, and most people never see most of them. I wanted one album that every guest adds to, where you can jump straight to the photos you’re in instead of scrolling through hundreds.

How it works

The backend is FastAPI on Azure Container Apps. Image files go to Azure Blob Storage, and the metadata used to search them goes to Azure PostgreSQL. Face clustering runs separately as an AWS Lambda job, so a burst of uploads after a group photo doesn’t slow the app down.

The main decision

Splitting storage this way was the decision that shaped the rest. Blob storage is cheap for large files but can’t answer “which photos is this person in?”, and a database is the reverse. Each one holds the part it’s good at.