SpendDrop

See where your money’s really going - drop, categorise, and conquer your expenses

Comment

Inspiration

I receive my credit card bill every month, and am often astonished by how much I spent! To really understand this, I needed to look deep into my transactions and figure out the trends.

There's currently two ways to do this: The manual way: manually note down all transactions in a spreadsheet and categorise them to figure out where I'm really spending - but that's a really laborious process that takes a lot of time.

The automated way: The other end of the spectrum is apps like Mint which connect to the bank directly, however a lot of times some of your banks are supported, and some aren't - there's always the worry about directly sharing bank credentials externally.

What if there was a better way - somewhere in the middle of these two ends of fully manual vs fully automated.

What it does

SpendDrop is the best of both worlds - it lets you only share what you want to share - a simple credit card PDF statement file that you get every month from your bank. SpendDrop then parses and extracts all transactions from the statement, categorises them using LLMs, and then shows you clever insights and charts like:

  • Spending by category (eg. Dining, Groceries, Shopping, Travel, etc)
  • Spending by merchant
  • Spending by day of the week
  • Spending by currency
  • And weekly spend patterns

How we built it

The app is a native SwiftUI app that renders charts within a ReactJS WebView. The data is parsed using a Firebase function with an LLM, and lots of PDF extraction code to support the wide variety of PDFs that banks around the world issue.

Challenges we ran into

  • Parsing PDFs: Due to the wide variety of PDF formats and tables issues by banks worldine, many in non-English languages, our PDF extraction is highly accurate and is trained to skip out on unnecessary bloat in the PDFs like legal texts.
  • Keeping your statement private: Your statement is not stored at all, it's processed in memory and never saved.
  • Keeping low cost vs high accuracy: LLMs are trained and tuned to provide highly accurate categorisation while still keeping cost low.

Accomplishments that we're proud of

  • SpendDrop is very accurate, and can parse a huge variety of statements from multiple banks like Citibank, Standard Chartered, AMEX , Chase, DBS. W've tested statements from multiple countries as well - US, Singapore, India, UAE and more.

What we learned

This is hard but very exciting problem to solve within the personal finance space. I'm excited to build SpendDrop as a perfect middle alternative between the fully manual or automatic way to tracking your expenses.

What's next for SpendDrop

SpendDrop will really shine once we integrate storage across multiple credit cards and statements for a user - it'll be able to show extremely valuable insights and trends over months and subsequently years, making it easier to see if your spends are where you expect them to be.

Built With

Share this project:

Updates