Advizr

Inspiration

Advizr was sparked from the common problem of getting accurate course information fast. The recommended way at universities is to go to academic advising and wait hours and hours until you get to ask a simple question that could be answered in a sentence. Furthermore, if you go to ChatGPT for academic advising for your courses, it will “hallucinate” non-existent courses. With a combined desire to solve a major student issue of waiting hours for academic help and optimizing an LLM to become a deterministic model, Advizr was born.

What it does

Advizr is a specialized academic advising RAG model that leverages Cohere’s API to provide users with accurate and reliable course information. By developing a RAG model trained on course data from universities nationwide, our model delivers precise, deterministic information on prerequisites, corequisites, grades, professors, and course descriptions. With its scalable architecture, Advizr has the potential to impact student academic planning and evolve into a comprehensive tool for career guidance!

How we built it

Before we started development, we conducted a survey that gathered over 100 opinions, revealing public interest in our idea. This motivated us to approach this opportunity with a solid tech stack composed of Python and React. We first performed data integration by pulling data from three different APIs: UBC Course Explorer, UBC Grades, and Rate My Professor. After that, we cleaned the data and, with the help of Cohere’s API, built a Retrieval-Augmented Generation (RAG) model for the purpose of providing accurate course data in an accessible form. Additionally, we used re-ranking algorithms by Cohere that led to an 85% increase in efficiency of the product. Once we had a solid backend, we began designing our vision of the User interface with Figma and making it tangible with React. To bring it all together, we used API calls to fetch data from the front end to the back end and vice versa. Hence, throughout this build process, we have created a convenient web application backed by AI that provides accurate and relevant information.

Challenges we ran into

Since we are new to the LLM space, we had trouble determining how to get started and fill our knowledge gaps. Another problem we faced was how to implement persistent data transfer between the front end and back end. Nevertheless, we were able to get past these obstacles by investigating relevant documentation, learning from mentors, and collaborating. This has resulted in the completion of our initial goal to create Advizr.

Accomplishments that we're proud of

Producing the first-ever academic assistant LLM web app powered by Cohere’s API in under 36 hours. During this hackathon, we have overcome various technical and physical challenges like lack of sleep and unconfigured virtual environments, all while learning complex technologies. In particular, we implemented an API we were not previously familiar with and built a RAG model for the first time.

What we learned

At Hack the North, we picked up a new tech stack and skills: Flask, React, Figma, RAG models and LLMs. However, not only did we build on our technical skills, we also grew our ability to collaboratively work towards a goal and problem-solve when blocked.

What's next for Advizr

We plan to scale our web app to include more personalized features like AI-powered course scheduling and student profiles for a more connected feeling. In addition, we aim to partner with universities from all around the world to help students feel confident about their academic journey by providing course information in an organized and convenient way.