Homepage
User provides input such as demographic and financial data.
Machine learning model predicts user's financial stress and overspending.
AI Agent provides educational lesson to the user based on their circumstance.
AI Agent provides educational lesson to the user based on their circumstance.

Project Story

Inspiration

Our inspiration was from a simple but one of the most important members of the society: students. In other words, it came from within. We always wondered how it is possible to save money given our loans, tuition costs, transportation, healthcare and the list keeps going on. We knew that financial planners existed, but clearly, we were still facing the same problems because it lacked a personal touch. We needed a person, more like an AI, that does the analyzing for us. We wanted something to solve basic problems like finances, to help us accomplish our long-term goals. Hence, the birth of HI-FI – Human Intelligence for Financial Independence.

What it does

Our app takes in user input such as age, gender, financial spendings, income and financial goals. This input is provided to a regression model which calculates the amount of money overspent by the student. Another classification model also calculates the student’s financial stress. These outputs from the machine learning models are passed into a Claude API to provide the student with educational lessons based on their financial habits to make them financially independent.

How we built it

Each component of the app was taken up based on expertise. Heraa worked on the ML model, sourcing, cleaning, and processing training data to create predictive and regressive models. Harsh and Pera explored Agentic AI, and wired Claude up into a FastAPI to be called by our React front end.

Challenges we ran into

For the machine learning model development, the original Student Spending dataset our team had found from Kaggle did not have the information required for meaningful predictions or analysis. Our team had to generate additional columns in the dataset using synthetic data from ChatGPT to ensure we got meaningful data relationships to analyze with the machine learning models. Our team was new to agentic LLM pipelines. We learned frameworks and flows from the ground up, and implemented them within 48 hours to create a comprehensive, cohesive, and safe flow.

Accomplishments that we're proud of

Our team is proud to have built this project within 48 hours for this hackathon. Both of our machine learning models achieved high performance and accuracy on our training and testing dataset. Our regression model predicted the amount of money overspent by students every month with a 0.95 R^2 and 3.35 MAE. Our classification model had a 98% accuracy and 99% F1 score. We are also proud of our emphasis on AI safety, implementing strict guardrails to prevent misuse and hallucinations in our client-facing agents.

What we learned

We learned frontend methodologies like React with TypeScript and Vite. We also learned how to integrate our ML model into Backend and how to connect everything up to the frontend.

What's next for HI-FI: Human Intelligence for Financial Independence

Our goal is to expand HI-FI’s platform to support millions of users across different regions, adapting recommendations for diverse financial systems, currencies, and cultural contexts. We can achieve this through partnering with tech companies and collaborating with like minded that have the passion to solve this problem to serve a bigger purpose Since our current modeI is based of students, we would like to partner with schools and universities. We can integrate HI-FI into student success programs, giving young people personalized financial guidance from their first year of post-secondary education onward.