StudyAI

Inspiration

The inspiration for Study AI came from the very common unsuccessful study group. Whether it be due to people not getting along or someone within the group being at a much more knowledgeable level, we wanted to create an environment for motivated students to easily develop successful partnerships with other classmates.

What it does

StudyAI is a website created to make the age-old question of Who do I partner with? a completely painless and very private undertaking. In the finished product, a first-time user simply inputs their name, contact information and transcript and an evolving machine learning algorithm instantly matches you with people you'd work well with, all the while eliminating any chance that another person will have to see your private transcript.

How we built it

The frontend website is run in Wix, which lets the user create a profile and input their transcript. From then the information is fetched and processed in a Google Cloud server, where the ML program is run to determine top matches. The information is then sent back in the 'Top hits' to frontend for the User to select from, or message. The ML algorithm (XGBoost model to eliminate unnecessary variables quickly) is built in Google Colab with semi-supervised learning. Since we had no training data whatsoever, we wrote a script to generate 1000s of synthetic transcripts via Gaussian probability distributions. Each transcript's 'strengths' in different subjects was independent. We then developed one more synthetic transcript to act as a human, and had the human select transcripts that would be an ideal study partner. The algorithm would then select a larger amount of transcripts and sort through to find the best ones given its training, and hand them to the human. In this manner man and machine worked together to develop the algorithm at an ever-increasing pace.

Challenges we ran into

Collectively, we had little experience with Full Stack development. Most of our time was spent learning and debugging. As a result, we never got the chance to set up communication between Wix and the ML program (the ML program itself had not finished its debugging stage). We also were not able to migrate our work to Google Cloud, and as such have a large Git repository.

Accomplishments that we're proud of

Creating the backend and extensive Git repo in such short of a time, as well as creating the synthetic transcripts and a way around the lack of experimental data usually necessary for ML programs. We also learned a lot in both web scripting and crawling. Although, StudyAI doesn't entirely work properly, we got some parts working with the course data and numbers. We left this hackathon becoming much more experienced than when we first started.

What we learned

Leaving this project, we all have a much larger understanding of the framework for website development, and the need for planning. As well as how to make everything communicates with one another. We learned a lot about how web scraping is used to pull information from any site and store those values to be used within the program. In our case, a script was made to pull every class name and number from the majors we used and store them in a list. Our experience with Python and Django grew dramatically as we found more ways to use these robust languages.

What's next for StudyAI

We'd like to fully finish the project, and get a decently wide userbase to develop the algorithm on. We'd also like to add in options to share things like books, or more complex social interactions.