Canv.ai | Devpost

Inspiration

Creating a solution that students can use to improve their academic journey is something that resonates with all of us. As students, we were all looking for a tool that would streamline the studying process, instead of finding one that did not meet our specific needs, we decided to make our own.

What it does

Through the Canvas API, we were able to gather all the files related to an active course. With this information, we created a RAG model that would give students the option to customize and tailor a study plan, summarize past lectures, and answer any burning questions. Unlike other generative AI models, our solution will give you answers based on the approaches and topics that your professor took to teach the class, with some Gemini ai assistance as needed; allowing for a more efficient learning process.

How we built it

Tech Stack: NextJS, TailwindCSS, MongoDB, Prisma, Python, Flask, Google Cloud. With the data provided from the Canvas API, we trained a RAG model on Google Cloud. This allow's the tool to give answers to specific questions on the answers.

Challenges we ran into

Our biggest challenge was getting the RAG model to work. We first started with Microsoft Azure to implement the RAG model, after some success, we eventually had to scrap that model due to problems we were not able to solve. We then pivoted to Google Cloud as we found it to be easier to deploy on their services. Another major challenge was getting the model to communicate with the frontend. Transcribing audio to text was also a challenge, but overcome with research and debugging.

Accomplishments that we're proud of

Working with and improving our RAG model to improve queries done on a user-by-user basis as well as carefully considered file conversions to RAG model readable files. The UI/UX is functional and adaptable, making it practical software.

What we learned

The most interesting thing we learnt was the actual RAG process. We found the concept of vectorizing both the data files and query and then using a generative AI model communicate with that was incredibly interesting.

What's next for Canv.ai

Our next goal for Canv.ai is to create a progress bar for all of the assignments that you have complete and give recommendations on what you should do next. .