Truist Coach: Introduction Image

UGAHacksX

Project by:

Kevin Xiang
Vaishali Prahalad
Mukil Sundaravadivel
Rohit Rao

Purpose

One of the most essential things in a user’s mobile banking application is their financial data. Of course, humans are visual beings — an influx of verbose textual data makes it difficult to analyze and interpret that data. With this in mind, we created our project, Truist Coach. Truist Coach enhances two aspects of mobile banking. One, it visualizes users’ financial data and financial history, allowing them to interact with the different proportions of their recent expenditures and the allocation of their money in various accounts and investment portfolios. Two, it uses a large-language model (LLM) to create hyper-personalized recommendations for how the user should approach the next month (in terms of budgeting, spending, saving, etc.) in order to increase their financial responsibility.

Tools utilized

For our front-end, we used React.JS to build a responsive web interface. This was connected to our backend using Flask. We used Python for the backend scripts - including training the model, generating and organizing data and connecting with OpenAI GPT-4 API for our LLM. We also used SQL to interact with a database.

Problems we ran into & how we overcame them

Learning new frameworks–we spent a lot of time learning how to start a React project, following which we had to enhance our frontend interface. We also had to learn prompt engineering and fine-tuning to successfully run an LLM on the backend that met the requirements of the project.
Another problem we encountered was creating links between the frontend and the backend. Although we were able to create an independent frontend and backend, we found it somewhat challenging to link them together. To overcome this problem, we spent quality time ensuring there was a strong link between the frontend and backend (i.e., users on the frontend can press buttons that call the model in the backend, returning a response that is then displayed on the frontend for the user to see).

Credit any public frameworks (APIs, etc.)

Used OpenAI GPT-4o API
Used open source react template licensed under MIT
- [(https://flatlogic.github.io/sofia-react-template/)]

Built With

Submitted to

UGAHacks X
- Winner Truist Hyper-personalization of Virtual Chat Assistant

Created by

I worked on the front-end, using React and CSS to create the chat interface, as well as linking to the back-end by using API calls to interface with our chatbot. Besides learning about React, I learned a lot about Git, and you could say that I learned by making mistakes, made evident by our commit trees!

Rohit Rao
I worked primarily on the back-end, creating and configuring the LLM to answer financial inquiries with personalized recommendation plans or actions. I learned a lot about Git & GitHub this weekend, as well as accessing and managing data in a SQL database, and how to fine-tune an LLM with specific prompt engineering.

Vaishali Prahalad
CS @ GT
My contributions focused on the backend. I interfaced with Google and OpenAI APIs and wrote detailed prompts to create personalized responses for our chatbot. I also constructed the Flask server for our frontend to connect with. I also worked on git branches so that I wouldn't have a git nightmare when someone else pushed to main!

Kevin Xiang
I worked primarily on the front-end, learning a lot about React development and data manipulation in the process. Besides from the Git commit tree nightmares we had, this project taught me a lot and was very rewarding!

Mukil Sundaravadivel

Updates

Rohit Rao started this project — Feb 09, 2025 02:29 AM EST

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.