ClearJar Capstone Project

My 4th year capstone project at UBC involved building an AI web application, focusing on delivering financial advice to users. We build this project for ClearJar, a startup company run by a recent UBC graduate. The goal was create a chatbot that can answer user specific questions about the user's finances. It tried to replicate the experience and advice someone would get by visiting their own financial assistant.

Stack was Next.js for front-end, Express.js for back-end, Pinecone vector database, PostgreSQL relation database.

There are many levels to this project. First we used a vector database to classify different types of transactional data based on semantic meaning. For example, when searching for data, a "coffee" and a "tea" would be close by in the 3D vector space. Then, we managed bank connections through Plaid, an external API service. Finally, we used Gemini, Google's AI model, to answer the user's question, based on data retrieved through plaid and Pinecone.

Our results weren't always consistent due to many factors, including our own algorithms and AI hallucination, but overall we had excellent results when interacting with our chatbot.

Here are some screenshots of the application.