FinAtlas

Inspiration

Our team members are only here today because our immigrant parents or grandparents made the decision to pursue a new life in the U.S. Some of the team members also grew up right here in Broward County. We recognize firsthand the immensely difficult path that immigrants put themselves on when deciding to move to a new country, so we sought to make one of the biggest hurdles--finances--more accessible. Our team believed we could leverage AI to connect users with local financial resources and institutions.

What it does

FinAtlas is a website where users can ask for help with financial resources (for example, which programs are available to low-income, single-parent households, or how can a broke college student apply for financial aid). Our AI-powered assistant then responds with specific, relevant, and, most importantly, _ clear and easy-to-understand _ suggestions and returns a map view of nearby financial resources that can help with the user's specific question.

How we built it

Our tool comprises two key components – an AI Assistant and Mapping Services. For FinAtlas's AI Assistant, our team leveraged Groq's free Llama3 generative AI model. First, on our main screen, the user is prompted to input a question. Behind the scenes, our backend executes a validity check to ensure the user is asking for help with financial aid/opportunities. If a question is not valid, the user is notified that their query is invalid. If a question _ is _ valid, our backend calls Groq's Llama3 model to research and return a response (the response is also scrutinized by rigorous validity checks to ensure it is relevant and accessible). Alongside a written response, our assistant returns nearby financial and economic locations in a map window on the page. This is achieved through a combination of Google Map's API and our assistant's searching capabilities. We also built a natural language processing model in order to execute our validity checks. However, the format of the output is inconsistent, so we didn't officially implement it into our backend.

Challenges we ran into

The first and biggest challenge we ran into was coming up with an idea. Our team wanted to work on a project that was innovative, practical, and to which we all connected. Many good potential ideas were discussed and picked apart, but throughout the whole brainstorming process we remained patient and honest with each other. After some time, we landed on FinAtlas, a project that would allow all of our members to play to their strengths and develop something we were proud of. Later, we found that trying to use OpenAI's API would be inaccessible to our team due to a paywall. We got around this by using Groq, a similar generative AI API that was free. Another one of the large challenges we faced was scope. Our team had many ideas we were excited to implement–evaluating local ratings, implementing machine learning models to assess the validity of queries, parsing through a large database of financial resources rather than receiving/generating it live to name a few. However, 36 hours is not a long time. Ultimately, we had to sacrifice some of the less crucial features to focus on our core mission of connecting users with local financial resources and institutions.

Accomplishments that we're proud of

This was half of our team's first hackathon and the other half of our team's second hackathon. We are immensely proud of our level of teamwork and communication that ensured smooth development. Our team is also proud of the numerous industry-level validity checks that occur to ensure the application does not hallucinate and stays relevant. This was also the first time one of our members used Figma for a large application.

What we learned

We learned a lot of new technologies Throughout this hackathon, we gained valuable experience with a variety of tools, both new and familiar, applying them in innovative ways to tackle our project challenges. We worked with platforms like Groq, Natural Language Toolkit (NLTK), Figma, and the Google Maps Platform, expanding our understanding of their capabilities. Additionally, we attended multiple workshops, including a Vanguard-led session and a Web Development workshop, which helped deepen our knowledge of CS networking and provided insights into leveraging new technologies in future hackathon projects.

What's next for FinAtlas

Though we weren't able to incorporate our natural language processing model, the model exists. We have plans to incorporate it for a stronger query validity check.

Built With

• canva
• css
• figma
• github
• google-maps
• groq
• html
• javascript
• natural-language-processing
• natural-language-tool-kit
• python
• react