Case study: Image repository

For a second experiment using an AI agent to build an application, I wanted something more ambitious than a basic app with text in a database. The goal was to create a system that would aggregate files from remote users, automatically pull them into a database, run configurable analysis tools on them, and make everything searchable and viewable with proper access controls.

The inspiration came from real life. I sifted through hundreds of photos to create slide shows for three family events this summer. Anyone who's done this knows the pain of searching through multiple directories to find relevant images. I was looking for a way that might make it easier, in the future, to find desired images of people and places.

So I asked Claude to build a photo management application. The core concept: pull in photographs from various sources, use AI-powered image recognition to automatically tag them (people, places, events, objects), and then make everything searchable and viewable through a clean interface.

For authentication, I stuck with what worked well in my first project—Google OAuth. I'm a firm believer in letting third-party services handle authentication whenever possible. It means I don't have to worry about securely storing credentials, managing password resets, or dealing with the security implications of holding user login data. Google handles all of that, and users get the convenience of single sign-on. Plus it's a good analog to real-world applications, where you typically authenticate against Active Directory.

The application breaks down into three main components:

1. A lightweight database interface for searching and displaying images. Users can search by tags, dates, people, or any combination of metadata.

2. A task scheduler that runs jobs at configurable intervals. This handles things like checking for new images, running analysis on unprocessed photos, and cleanup tasks.

3. Integration with external AI tools for image analysis. When new photos come in, the system queues them for processing, runs the analysis, and stores the resulting tags and metadata. By making it external, it allows for upgrading when better AI tools are available. The current script-- the stag.py program-- can identify types of images but not specific people. The end goal is to be able to group images by identification of specific people in them.

The whole thing came together faster than I expected. Having learned from my recipe database project about working with Claude in smaller, iterative chunks, this build went more smoothly.

The code is available at: https://github.com/davesnyd/PhotoSort-V1

• 

• 

• 

•