Pufferfish

Inspiration

As young adults transitioning into the broader financial system, we've all started to realize just how complicated it gets. To make the most of your money, and do so in the safest way, there's complex setups of different accounts and assets, and ensuring your money is safe can require daily monitoring. However, for many, this simply isn't something they want to and can do. There are existing tools out there for fraud detection, such as CreditKarma or RocketMoney, but they focus on those who are very integrated into the financial system, having multiple accounts and credit cards. For those who are underbanked, however, the complexity of these tools and the sales pitches they fold in simply are too much hassle. As a result, we decided to set out to create a simpler, friendlier tool to protect users from and inform users about fraudulent transactions, without all of the complexity and upsells.

What it does

Pufferfish is a mobile application that helps you monitor the transactions on your bank account for various common types of fraud, such as:

Double-charging (a vendor charging you twice, hoping you won't notice)
Suspicious transaction destinations
Large transactions on peer-to-peer payment services (like Venmo and Zelle)

The app alerts users to potential fraud instances, as well as providing a friendly AI-generated summary of the incident.

Additionally, it provides other visualization of a user's transactions, such as by time of day, allowing them to easily visually identify anomalies.

How we built it

The front-end application is built using React Native. The back-end API is implemented using FastAPI, using MongoDB Atlas as a database. Financial data is brought into the application using SimpleFIN Bridge, and Samba Nova's API was used for LLM-based analysis and summarization.

The API is dockerized through a Github Action, and hosted on DigitalOcean's App Platform. The DigitalOcean App Platform app and MongoDB Atlas database are deployed via Terraform, continuously integrated using the HCP Terraform platform.

Challenges we ran into

Difficulties with CORS and data exchange between frontend and backend
Balancing ambition and the constraint of 24 hours to develop
Spotty internet in the venue

Accomplishments that we're proud of

First time working on a real React Native application
3 first-time hackers in the group
Fully continuously integrated backend and infrastructure, ready to continue development and scale up

What we learned

How to create and work with complex REST APIs
How to implement authentication for an API
How to prompt engineer in order to get programmatically valid output from an LLM
How to use DigitalOcean's App Platform for simple deployment
Using Github and HCP Terraform to continuously integrate a repo of Terraform IaC
How to use MongoDB Atlas

What's next for Pufferfish

Further feature implementation and polishing of the frontend
Providing more interfaces for financial data on the backend
Implementing further scalability on the infrastructure, such as adding load balancing and redundancy