Study Sphere

Inspiration

I was a tutor at Mathnasium for all of highschool and during our session we were able to help our students with their homework. A couple others worked at Kumon and had tutoring experiences at school. But the issue is, We were not in their classes and we don't know what exactly their teacher wants and doesn't. As tutors, we don't have access to their notes and have to adapt to whatever is thrown at us which induces a lot of human error on the tutors end. Additionally at home we always helped out our siblings with their homework and spent hours creating practice tests for them based off their notes to assess their content knowledge. Thats when we realized using AI to directly interact with your notes and automatically generate tests is the best way to help students study. StudySpheres testing accuracy would be way better than any tutor and can provide results for students right away and is FREE.

What it does

StudySphere connects directly to your notes and analyzes them allowing you to not only chat with your notes but also create practice tests for you to better understand your notes. It takes in many different types of file formates including

How we built it

Our front end is built in solid. Our backend is separated into two parts, one intermediary backend and the other to be the main backend, called Arguflow. The backend is a rust actix web server with postgres qdrant as the primary storage mechanisms. We use open. AI for embeddings as well as chat.

Challenges we ran into

We ran into a couple of challenges throughout the project. At first we wanted to create a chrome extension but it was too difficult to develop in the time we have because there were too many AI's to consider. Another challenge we faced was that when we first created the front end using the sveltkit we realized it was hard to hook it up with the backend in a reasonable amount of time. Working with the google drive API because of its complexity, especially the file picker mechanism.

Accomplishments that we're proud of

We are proud of how seamlessly StudySphere hooks into the notes. We are also proud of our future expansion ideas, the fact that we can create this educational app into a social media platform for education.

What we learned

We learned how the hackathon process worked, as we are freshman so this was our first experience. We also learned how sensitive LLMs are to different prompts. We also learned about OAuth, and message passing works between frontend and backend.

What's next for Study Sphere

StudySphere right now would be a tool marketed towards college students but we can expand to a greater student population by partnering directly with schools and university. Through these partnerships we can create libraries of consented student notes where users can pull from to generate their tests. Additionally we can connect directly with whatever platform the students are using at school(ei canvas) to make our bot, SIA, have access to analyze school provided files and slides. We could also partner up with Online courses and google education, this partnership would allow these online courses to use our tech to understand the retention of their students and use the data to continue improving their courses. Additionally, our tech can be used beyond edtech, as SIA can sort through vast files of data and pull out exactly what you need which any large company can benefit from. In the future we are planning on implementing a review option after each test generated to train the model to create more accurate tests. We also plan on advancing to GPT 4, as we are using GPT3.5 right now. But we would only implement GPT 4 to be used for test generation as GPT3.5 does a pretty good job with analyzing notes and chatting with them. This way we could make our model cost effective.

Built With

• flask
• google
• javascript
• openai
• postgresql
• python
• qdrant