Card Sense

Inspiration

The inspiration for our project comes from the challenge of navigating financial decisions, particularly when choosing a credit card or business financing option. With hundreds of choices—each offering different rewards, interest rates, and benefits—it’s overwhelming to find the best fit. Our goal is to simplify this process by using AI to provide clear, personalized recommendations based on user needs. The best software eliminates complexity, and that’s exactly what we aim to do.

What it does

CardSense is an AI-powered financial assistant designed to help users navigate complex financial decisions, including credit card selection, loans, and financial literacy. By web scraping data from over 100 credit cards and financial products, we provide personalized recommendations based on users' financial goals and spending habits. Our AI-driven approach simplifies overwhelming financial choices, ensuring users can make informed decisions with confidence.

How we built it

First, we built a web scraper to gather data on credit cards, loans, and financial literacy topics, including names, perks, rewards, interest rates, and credit-level requirements. We then embedded this information into a vector database and leveraged Retrieval-Augmented Generation (RAG) to improve the relevance of search queries. Finally, we developed a Flask backend and a React frontend to provide an intuitive interface where users can query the LLM for personalized financial insights.

Challenges we ran into

Compiling the data was the first issue. The website we scrapped had inconsistent HTML class names and a difficult UI making forcing us to get creative with how we structured our code. Next, we spent a while insuring the LLM was giving accurate and relevant responses.

Accomplishments that we're proud of

Our first challenge was compiling the data. The website we scraped had inconsistent HTML class names and a complex UI, requiring us to structure our code creatively to extract accurate information. Additionally, ensuring the LLM provided relevant and precise responses took considerable effort, as we fine-tuned the retrieval process to improve accuracy and eliminate irrelevant outputs.

What we learned

One major challenge was balancing AI accuracy with real-world financial constraints, ensuring that recommendations were both factually correct and genuinely practical. Additionally, we had to navigate the complexity of processing unstructured financial data, as credit card terms vary significantly across issuers, requiring robust parsing and intelligent data structuring to maintain accuracy and relevance.

What's next for CardSense

There’s a lot we’d like to improve. For starters, we plan to implement Firebase Authentication and Firestore to enable persistent chat history, allowing users to track past recommendations. Additionally, we aim to refine the LLM’s capabilities, enhancing its ability to provide more nuanced financial insights and expanding its functionality for a broader range of financial queries.

Built With

• beautiful-soup
• chromadb
• flask
• hugging-face
• langchain
• openai
• python
• react
• selenium
• typescript