-
FinanceGPT is capable of understanding slurred and improper user inputs.
-
Standard Functionality
-
Application Design
Inspiration
Open first arriving at the hackathon, our team was unsure of what to build. We knew we wanted to make something of great utility and that would be able to service a large community. However, after much brainstorming we couldn't quite land on a single solid idea. But after attending the Microsoft workshop for Semantic Kernel and speaking with some of our peers/competitors we got the idea make an AI/LLM that would be able to help its user optimize their spending and saving to get the most of their money.
What it does
FinanceGPT is an AI chatbot designed to be able to provide financial advisement services for users. The system was designed to operate like an extension or add on for a banking service. In which, FinanceGPT would be given access to some of the user's financial data to then be able run an analysis on and make recommendations to the user about how they could better spend, save, and manage their money.
How we built it
Our project was developed entirely in Python and hosted on a GitHub. All team members used Visual Studio Code to allow for a streamlined development process. The backbone of our project is Microsoft's Semantic Kernel. We used this API to allow us to integrate OpenAI's ChatGPT-3.5 LLM into our system and modify it to meet our vision for the project. For the front-end, we used PySimpleGUI to make the application window to display our service as opposed to running it straight out of terminal. Minor notes include the use of JSONs for persistent data storage.
Challenges we ran into
By far the most difficult challenge we faced was designing a functional graphical user interface. No one out team had any prior experience with creating GUIs, and that fact is certainly exhibited by the look of our final product. However, despite our unfamiliarity, we as a team did not allow the constant errors and bug to weaken our resolve and determination.
Accomplishments that we're proud of
This being the first hackathon for all of our team members, we are proud to say that we even competed in such an event. This opportunity has been exceedingly wonderful. We have each learned an immense amount and expanded our knowledge bases and professional and social networks.
What we learned
Having never made a GUI before, everyone on our team learned a great deal about how to plan, design, develop, and integrate them. Additionally, working the Semantic Kernel and OpenAI has deepened our understanding and appreciation of LLM's and we each expect and look forward to leveraging similar technologies in our future projects.
What's next for FinanceGPT
If we were to continue to develop and expand upon this project, one of the first things we would do is improve the GUI. Being our weakest element, there is much to improve upon. We'd likely seek to move to a web application and try out a JavaScript based front end. And along with the move to a web-based system, we would also like to work on a plaid integrate so that users would be able to connect their bank accounts directly to the service and not have to rely on whether or not their financial institution has integrated supports systems of their own. Lastly, but certainly not least, we would want to fine tune and train our LLM to be able to offer even higher level of service to users.
Log in or sign up for Devpost to join the conversation.