I built a personal travel app with Claude

So, yesterday I made a post about noticing many users complaining about Fable being nerfed and how I thought it must be something people are doing differently to not get the results they were looking for. Someone made a comment that lots of people often say they either get Claude to work well or it’s rubbish but never explain what it is they’re working on. I thought I’d share one of my personal projects.

This is a personal app I’ve built for myself, called Live Rail. Simplistic, I know. I live in the UK and regularly get the train to London twice a week. While we have an abundance of rail related apps, I was frustrated by one problem. Stations love to hide platforms numbers until 15 minutes ahead of boarding. For small stations, this works but if you’re at London Euston every week like I am (UK Redditors will know what I mean), it cause chaos because of the sheer amount of people queuing up and suddenly making a dash when the platform finally reveals itself. Like a sick game of musical chairs.

Now, there are services out there that will attempt to predict the platform of trains and I’ve tried those but I wanted something in an app form and thus my idea to build this was born. It’s a fairly simple departure and arrival app but the key features I wanted for the backend was a predictive platform engine and a live tracking feature so I can see where a train is on route and if it’s making good time.

The UI still needs some work but might not get around to it as intend to use for me personally but the backend consists of Network Rail APIs providing all of the information you see here. I’ve built an adapter that takes their APIs and morphs it into information that I wanted in the app by producing new APIs that live in the adapter itself.

The most difficult part of this was getting the predictive engine to work and while it still needs work. It’s normally right about 75% of the time. The engine works by checking how many times a given train has arrived at a platform for a station at the end of a journey and compares it to the advertised platform number that network rail provided when it released a schedule. I save these comparisons in a database and over time it builds up and the engine makes a prediction based on the confidence levels that each train journey accumulates.

But keeping this light, all of this was 100% built with Claude. About 80% Opus 4.6 and 20% Fable 5. This was not a one shot or a few prompts. It took a few sessions to get the UI and backend up and running. It will take a few more to refine the predictive engine but for a personal app. I’m quite happy with how its turned out so far.