StudyBuddy.io

Inspiration

As university students, we often felt the lack of meaningful feedback in our learning journey. This inspired us to create an AI-powered platform that empowers students with free, instant access to detailed insights and progress tracking. Our goal is to help learners grow, improve, and become better every day.

What it does

Take your graded exam PDF's and produce and leverage LLM models like chatGPT, LLAMA etc to generate a detailed feedback into your topic of strengths , weak topics and a short blurb into your next steps.
Generate dynamic time based mini tests leveraging LLM technologies for a quick practice session for topics of your choice
Efficiently track your progress and improvement based on your performance in graded exams that you as a user upload.

How we built it

Designed and annotated a synthetic dataset using Cohere's text generation model, CVAT annotation tool for the purpose of training our own Token Classification/ Question Answer classification model.
Passed extracted Question Answer results from the model to Feedback pipeline which integrates LLAMA, GPT-4o and MISTRAL models to generate the best feedback for the user as well as accurately evaluate user's responses to the questions in the document
Dynamically generated time based Mini tests for users leveraging the same model technologies as described above based on user topic selection.

Challenges we ran into

Determining the best approach to design our data extraction and feedback pipeline which are the key highlights of our project.
Integrating the use of LLM technologies efficiently to provide constructive results to the user in as limited period of time as possible
Designing an intuitive yet component heavy interface to provide the best services to the user.
Connecting to Postgres database locally and running docker.
We unfortunately couldn't successfully integrate the feedback and data extraction pipeline due to payload issues

Accomplishments that we're proud of

Great teamwork and planning into the flow of our website
Successfully developing a low inference time based data extraction and feedback pipeline
Seamless and intuitive UI for the user.
Learning about new technologies and different model approaches to tackle real world problems

What we learned

Payload can be a big issue so look into that in advance
Integrating backend and frontend is a humbling task
Design and workflows and constant communication are key to successful development
Timezone differences don't matter when it comes to collaboration