-
-
icon
-
screenshot
Inspiration
My parents and I are like many in Brazil: we mix personal and business money together.
We tried using discipline, separate bank accounts for business and personal use. Eventually money flew between accounts again. Convenience and urgency beat us.
We tried using existing financial control apps, but they all either:
A. Ask that the user always manually register expenses (which isn't sustainable) or; B. Try to automate classification, with tools like ML and LLMs, which doesn't work because brazilians name their parking lot business "João da Silva Bar" or their hot dog stand "Rogério". People change what the business does and don't update metadata critical for any automated tool.
Only who spent the money knows well enough the context (for now).
Mudda is the solution.
What it does
Mudda leverages brazil's open finance to get as much data as possible and not require that the user stop their busy day to register the expense.
Still, the expense needs to be classified so we leveraged radial menu used in first-person-shooters to make classification a 3 second or less activity.
Then we present the data on a readable and intuitive format for users to take control and tackle the expenses that matter the most.
How we built it
We use react and react-native, mostly, along with LLMs.
Challenges we ran into
By far, the biggest problem is the fact that most financial institutions don't actually adhere fully to OpenFinance standards, so we had to collect lots of data and activity to learn the ways in which they fail and work around them.
Some institutions simply don't report activity, or report with wrong or missing fields. Mudda's projector takes those messy events and handles them.
Accomplishments that we're proud of
We have a projector that knows the various ways in which financial instituions mess up in credit, debit cards, loans, PIX transfers, investments and works around them.
What we learned
Financial data is dirty.
What's next for mudda
Some competitor apps try to just take data as it comes raw from OpenFinance and slap an llm tasked with analyzing the mess. However as we stated before there is simply no way anyone (including an llm) can know that "Maria da Silva, 35 BRL" was money spent to change your car's tire. The data has to be clean for such rich analysis.
Ultimately, mudda is a data curation tool: the user curates and organize their data. Curated financial data is a treasure trove. With a mature product, we can later add llm analysis that produces actually useful artifacts and specialized UIs for the user to study their own behaviors.
Log in or sign up for Devpost to join the conversation.