Inspiration
Always having wanted to track my expenses but never finding a good application to do so. Simple bank-transaction based budget trackers miss a lot of important detail, but inputting everything manually is not sustainable unless someone has insane discipline.
What it does
You can take pictures of your receipts an have each item on them be categorized one by one by a custom ML algorithm, then be able to see in great detail where and when you spend too much money. Some people spend too much because they drink too many beers on Fridays, some drink too much coffee on weekdays. This app reveals it with little effort.
How I built it
I built it using Swift for the best integration and Apple's great API's, and trained the classification model in Python with PyTorch.
Challenges I ran into
Tons. Being able to extract information from receipts without overhead costs, achieving a good enough accuracy, creating a fast enough scanning pipeline, designing an inviting but detail-rich UI.
Accomplishments that I'm proud of
The design looks much better than I initially thought I can put together, and the model also runs very fast, very accurately. Early feedback suggests the app has a niche.
What I learned
How to use Swift, how to train complex neural networks, how to publish an app, how to design an application, how to conduct user testing, how to scrape data, and much more
What's next for RCT
For now its only works properly in the Netherlands, so I want to make it really good for this market. I'm currently reworking the categories to not overwhelm users with too much information, then I'll launch a bigger marketing campaign and start a/b testing like crazy. Afterwards, making the app work well in other countries too will be the next step, first and foremost the German speaking countries.
Extra info
The screenshots attached might differ a bit from the app's state in the app store if Apple has not yet approved the big category redesign.
Log in or sign up for Devpost to join the conversation.