ZoodleMe

This is the project I have worked on the longest, and also the one where I was worst at my job when I started.

The first version of the backend was written with a ChatGPT tab open next to the editor, genuinely learning what an API was while building one. AWS in particular felt hostile in a way I did not yet have the vocabulary to argue with. Deploying a single Lambda function took the better part of a week the first time, mostly spent finding the one permission I had forgotten in a policy I did not understand. The same deploy takes under ten minutes now, not because Lambda got easier but because I stopped treating infrastructure as something to be feared into working and started reading the documentation properly.

What it actually does

ZoodleMe generates stories for children with an LLM, narrates them with text-to-speech, and illustrates them through an image pipeline, so a child gets something closer to a finished picture book than a wall of generated text. Payments run through four providers in production — Stripe, Razorpay, and Apple and Google in-app purchases — because a children’s app has to support however a parent’s phone happens to be set up, not the payment method that is easiest to integrate first.

The estate it grew up on

ZoodleMe and Edactly share a single serverless AWS platform: FastAPI on Lambda, RDS PostgreSQL, Cognito for auth, all seven Lambdas and the API Gateway defined in Terraform with OIDC deploys. ZoodleMe came first, and a lot of what makes that shared estate solid now exists because it broke on ZoodleMe first and got fixed before Edactly had to hit the same thing.

Where it stands

1,600 users, real payments processed across all four providers, and a backend that has been rewritten enough times that almost none of that first ChatGPT-assisted version is still running. That gap, between the code that taught me what an API was and the code running today, is the clearest evidence I have of how much has actually changed since I started.