← All Projects

TV Planner

A personal TV planner to keep track of programmes across streaming services.

I subscribe to several streaming services, and although each has its own watchlist, there's no easy way to manage everything I want to watch across them. I wanted one place to keep track of programmes I was interested in, what I'd already watched and which series had new seasons coming out, so I could decide what to watch next without jumping between services.

I'd had the idea of building an app to do this, and while I was on holiday I decided to give it a go. I'd been experimenting with AI-assisted development, so I was curious to see how far I could get using ChatGPT and Claude through their mobile apps, without even sitting down at my computer.

I'd originally imagined a proper mobile app, and that's still something I'd like to build. But I thought it made more sense to get the functionality working as a web application first. I had a reasonable idea of what I wanted it to do, although I deliberately didn't plan every detail in advance. Part of the fun was seeing what I could build with AI and developing the idea as I went along.

Product screens and details

Two phone screens of TV Planner: the catalogue list showing the service and release for each programme with a Save button, and a search for detective with poster artwork Two phone screens of TV Planner: the planner with one programme under Up next and four under Want to watch, and the lower part of the planner with the empty Maybe and Watched sections

Catalogue

Search tens of thousands of programmes, start with the networks and services I use and filter by newest, upcoming or A to Z.

Planner

Sort what I have saved into Watching now, Up next, Want to watch, Maybe and Watched. The planner stays on my phone.

Select a screen to enlarge it. The screens show the live app with a few programmes saved.

The story behind the project

The product

TV Planner is a web application for keeping track of television programmes across different streaming services. I can browse the catalogue, save programmes to a watchlist, keep track of seasons and mark what I've watched. The idea is to make it easier to find something worth watching without having to remember which service it's on or keep separate lists.

I wanted it to work comfortably on a phone, including when I'm offline. It's a progressive web app, so it can be added to the home screen without going through an app store. The shared programme information comes from TVmaze and is served from a Cloudflare catalogue, while my own watchlist, ratings, notes and viewing progress stay in the browser on my device. That means I don't need an account just to use it.

The catalogue is considerably larger than when I started. It now covers tens of thousands of series and miniseries, rather than the few hundred programmes in the original version. That gives me more to choose from, but it has also made search, filtering and deciding what to show on a small screen much more important.

Building it with AI

The first version came together while I was on holiday, using ChatGPT and Claude from my phone. I was interested in whether I could take an idea from a conversation to a working application without sitting at my desk. It was a useful experiment, but I also found myself making quite a few product decisions as the build progressed.

Initially the catalogue was relatively small, with information gathered from Wikipedia-based lists and enriched through TVmaze. As the project developed, I started questioning why I was maintaining a limited collection when TVmaze had a much wider range of programmes available. I decided to move away from the original discovery process and build a proper shared catalogue instead.

That turned into a much bigger piece of work than I'd anticipated. In September I worked through a migration from 491 programmes to more than 34,000, using TVmaze as the main source. We had to decide which programmes qualified, deal with ambiguous titles, preserve existing records and make sure catalogue changes didn't interfere with anyone's personal watchlist. I wanted to include a broad range of series rather than just a handful of popular streaming originals, but that meant accepting some gaps and imperfections in the source data.

I also made a deliberate distinction between catalogue information and personal viewing data. The shared catalogue lives in Cloudflare D1 and can be updated or repaired. Personal choices, notes and viewing progress stay in IndexedDB on the device. If something goes wrong with a catalogue import, it shouldn't wipe the list of things I want to watch.

As the catalogue grew, I realised the more difficult problem wasn't getting the data into the system but making it useful. Thousands of titles are no good if browsing them on a phone is a chore. I've spent time revisiting the navigation, programme artwork, filters, search and how I distinguish between things I want to watch and things I've already seen. Some ideas that sounded useful at the beginning, such as additional screens and viewing states, ended up creating more complexity than they were worth, so I've also been simplifying the interface.

The application has changed direction several times as I've learned more about what works. I still want to turn it into a proper mobile app eventually, but I would rather get the core experience right before taking on that next step.

Results and reflections

TV Planner is live, with a much larger catalogue and more functionality than the original experiment. I've learned a lot about building with AI from a phone, working with external data, moving from a small dataset to a much larger one and managing the difference between shared information and private user data.

I'm not going to claim it's finished. The catalogue work has progressed further than the interface, and I still need to spend more time using it properly on my phone to decide whether the planning experience does what I originally wanted. There are also improvements to catalogue updates and distribution that I haven't completed.

Looking back, the experiment wasn't just whether ChatGPT and Claude could build an application from a phone. They could help me get surprisingly far. The harder part was knowing what the application should do, keeping it straightforward as more possibilities appeared and resisting the temptation to treat additional features or more data as progress in themselves.

Have an idea or a process you want working?

I take ideas from spec to a working product, and the same approach applies to automation and AI projects inside organisations.

Contact