- Robert Conde
- Nikhil Iyer
- Grayson Kornberg
- Jonathan Mesa
- Rebecca Weinstein
A recommendation system to suggest smartphones by evaluating the user's purchasing behaviors, preferences and interaction history with the chatbot. Created for the University of Florida Gator AI Hackathon.
Single page application with smooth and intuitive UI primarily built using React, Bootstrap, and Vite.
Implemented using Flask and to facilitate API calls to OpenAI's state-of-the-art large language models. Vivian's responses are facilitated using GPT-4, and was granted the ability to make direct requests for information from a large set of structured data. This enables Vivian to give responses with specific, accurate, and up-to-date information
We collected current data from Verizon's selection of available smartphones keeping track of information such as price, color, storage, camera, battery, images, URLs, etc. Data normalization and semantic embedding was needed to remain robust against the variety of smartphone formats. This allowed us to search our structured data using unstructured queries taken directly from Vivian. This embedding strategy also allowed Vivian to draw graphical elements to the frontend.
