#6 - University Blue

AI Chatbot for Questbridge applications which provides answers about all Questbridge partner schools, from CSS and FAFSA codes to student culture to financial aid information.

Comment

How we came up with Questbridge AI

Background

We were inspired to create this chatbot specifically for Questbridge applicants because we wanted to provide access to higher education for those who need it the most.

One of our team members currently mentors low-income and high-achieving high school students through their college process. They noticed the same questions would always be asked, but there was a shortage of trustworthy resources and easy application websites to navigate to find those answers. Students would often not have the time to scour each and every single page, leading to unnecessary frustration and crushed dreams. This is why we created QuestBridge AI, so we can bring quick and credible answers to break these barriers.

What it Does

Questbridge AI is an all-in-one admissions guide for students interested in the Questbridge program, which aims to help under-resourced and first-generation college students attend colleges throughout the United States.

Questbridge partners with many universities, including MIT, Yale, and Cornell. It's historically helped students obtain full-ride admissions to prestigious universities, with a network of over 25,000 alumni and current scholars.

How we built it

Questbridge AI would not be possible without the help of Microsoft Azure and Cloudforce, which provide the architecture supporting the chatbot.

The data for this chatbot came from the Questbridge website, which lists information relevant for each and every college within the Questbridge program, such as FAFSA number and application deadline. The data was scraped with Selenium, which crawled through the Questbridge website to download all relevant data into a JSON format.

After the data was scraped, we parsed the JSON files into a .txt format with Python, then loaded the data into Azure Cognitive Search. Using ACS, we were able to extract relevant information from the documents to inform a GPT 3.5-turbo model of up-to-date information regarding Questbridge applications for potential Questbridge scholars.

To display our model, we created a React application running on Vite. To style our application at lightning-fast speed, we used Tailwind for CSS, and deployed the site using Netlify.

Challenges we ran into

Our biggest challenge was with API architecture. Neither of us had extensive experience with REST APIs, and while it was very quick to scrape the Questbridge website - we finished by the end of Day 1 - we took a large portion of Day 2 figuring out the Azure API, but once we figured this out we were able to rapidly set up and deploy our site thanks to Tailwind, Netlify, and Vite.

Accomplishments we're proud of

Genuinely, we think that the thing we're proudest of is the fact that we built an application which can have a positive impact on the lives of those who apply to college through the Questbridge program. Questbridge scholars have it hardest when it comes to college applications - lack of resources and maintaining a greater amount of responsibilities to themselves and their families puts many students at a disadvantage when it comes to college applications because they do not have the luxury of scouring for the information they need. We hope that this tool can help level the playing field.

What we learned

We learned a few important things:

  • The importance of knowing how to operate REST APIs well
  • AI architecture and how AI architecture can work in the context of domain-specific information

What's next for Questbridge AI?

Going forward, we have more ideas to make Questbridge AI even more comprehensive. One important improvement would be to web scrape from even more websites and gather further data to train our chatbot. Ideally, QuestBridge AI could be expanded to include universities beyond Questbridge as well, providing a fully personalized and accessible experience.

Given more time, we would also have liked to tailor our chatbot further. Fine-tuning the parameters would allow us to have the chatbot give less redundant answers and more specific answers to better serve our students.

Share this project:

Updates