-
-
Fiscora Front Page
Inspiration
Have you ever entered a store, Target, Abercrombie, or even Louis Vuitton with the intention of not buying something but ended up taking a significant toll on your wallet? Well, that is a common struggle, and especially as college students who have to suffer the torment of nearly inedible dining hall meals, it is seemingly imperative to practice conscious spending, though it can seem daunting. However, it doesn’t always have to be. Contrary to its connotation, conscious spending introduces an optimal balance between personal spending, investing, and savings. Listening to the idea made it appear rather simple to implement in our lifestyle; however, when it came to actually implementing it, it seemed nearly impossible without a source of motivation. This is where Fiscora comes in, your personal financial friend.
What it does
Fiscora takes a user’s banking info and performs a multitude of deep analyses on it in order to provide the user with a full perspective on their spending as well as a positive alternative to impulsive purchasing. Fiscora’s main features include data visualization, where it maps the user’s spending into four categories on a pie chart, allowing the user to see where exactly their money is going. It also utilizes generative AI to analyze individual transactions in order to understand the user’s spending patterns and see where overspending can be reduced. From this data, the LLM provides constructive feedback on how to cut down in certain areas, and a specific amount of money that could be saved by doing so. This number is then put through our S&P 500 model, which outlines the net positive that a user can make by investing that money instead of spending it. Our goal with Fiscora is to show users these productive financial decisions they could be making.
How we built it
As we intended to base our analysis on an individual's bank transactions, we needed an API that would provide that functionality, and Plaid did much more by providing a tool, Sandbox, allowing us to access simulated bank transactions, optimal for development. We had initially developed a dummy application that pops up Plaid’s Link UI, which we kept separate from the frontend in charge of visually displaying our analysis. Secondly, we wanted to challenge ourselves, and thus developed our own deep neural networks using Keras/Tensorflow to model trajectories of S&P 500, and thus the projected growth of investments if made now. We additionally utilized Google Gemini to provide live constructive feedback based on the bank transactions, along with personalized encouragement to initiate action.
Challenges we ran into
For both of us, LA Hacks was our first experience with most of the technologies we used, such as Flask, React, Vite, and TailwindCSS. Naturally, figuring out how to work with these languages came with a learning curve, whether it be making the GET and fetch requests to interact between the frontend and backend, hiding API keys within Github, or understanding the imports necessary to run a file, every line of code required intention and detail. One issue we ran into on the frontend side was with formatting the data that the Google Gemini API responded with when queried. Understanding how jsonify() worked in order to transfer data between the front and backend was also a crucial part of our learning process.
Accomplishments that we're proud of
Overall, we’re both proud to say we came out of LA Hacks with a finished, working product! As we said, working with completely new technologies was quite a challenge, yet we managed to create an MVP within 36 hours, integrating new languages, APIs, a deep neural network, and LLMs.
Arshia: For this project, I learned how to use React, Vite, and TailwindCSS in order to improve the design and overall functionality of our product! I learned a lot about how to pass data through the front and backend, and how to format this data in a way that constant functions in the frontend could understand it, as well as dynamic backend programs. It was also my first time integrating an API like Plaid into a project and collecting and parsing through all its data. Every technology we used had its own learning curve, and I can definitely say I finished this hackathon with the ability to approach new languages and technologies confidently.
Jisha: I learned how to develop deep neural networks along with gaining a conceptual understanding of the union of backend and frontend. Moreover, this project taught me the importance of expanding the scope of what you initially intended to do. Specifically, we had initially planned on only implementing two features (a personalized plan and visual data representation); however, we later sought opportunities to tangibly help users achieve their tasks (motivation with positive reinforcement/ visual representations of possible success)
What we learned
As beginners to the hackathon environment, we learned a lot about the process of ideation, planning, and development when it comes to software products. One thing we struggled with in the beginning was debating on an idea that would be the most appealing to consumers, judges, and also ourselves. However, we missed the main point of many software applications – their true audience. Through our struggles in the ideation stage, we realized that the most important part of formulating a project idea is to focus on the people you’re impacting and the scale of that impact.
What's next for Fiscora
The possibilities for Fiscora are endless; We have built a foundation with four primary forms of analysis/encouragement, however, there is scope for many more, utilizing agents to specialize in tasks and interact amongst themselves to provide the most optimal user experience. Additionally, we intend to save users’ info in a database, allowing users to leverage the functionality in the long term (ex., regularly sending personalized messages, updated analysis strategies).
Log in or sign up for Devpost to join the conversation.