These tablets left the factory with the apps already installed. If something broke later, getting to the device wasn't exactly easy.

These were educational tablets for students.

The apps were preinstalled by the manufacturer before the tablets were boxed and shipped.

They also had to work offline.

Not "the internet might be slow" offline.

Actually offline.

The only time a device needed a connection was during activation. After that, it might never connect to the internet again.

That meant the app couldn't depend on an API, a cloud database or downloading something when a student needed it.

The books, media, app logic and everything else the student needed had to already be on the device.

And because updates were manual, I couldn't assume I'd be able to push a quick fix whenever something went wrong.

The source material came as ePub files from the client.

I couldn't just put them inside a generic ePub reader because the app needed to understand the structure of the books.

I had to extract things like titles, subtitles, chapters, links and other content and turn them into data the app could navigate and index.

Some of the books also contained video.

The video itself didn't need to work offline, but I didn't want students to leave the reading experience just to watch it.

So videos played inside the app rather than opening a browser, YouTube or another player.

The same idea applied to the other media in the books.

Audio, PDFs and other assets were handled as part of the reading experience, with the content stored locally and decrypted when it was needed.

The tablets were going to children, and they weren't going to have someone sitting beside them monitoring what they did on the device.

So the suite included its own browser with restrictions on what students could access.

A lot of the work here was testing.

I spent time trying to get around the restrictions myself — opening different types of sites, following redirects and trying the obvious things a teenager would try.

Whenever I found a way through, I closed it.

It wasn't particularly glamorous work, but it mattered more than making the browser look good.

I joined the project fairly late.

The manufacturer had already finished its part and was waiting for the software, so there was a fixed hardware production schedule on the other side.

There were also several apps to finish.

The target was roughly two months for the whole suite.

That meant I had to make decisions quickly without creating a codebase that would become difficult to change later.