EffiSTEM

EffiSTEM is a customizable and efficient learning companion which leverages LLMs and STOTA STEM OCR Mathpix for providing high quality answers based on your lecture notes and Youtube playlists.

Comment

Inspiration

Large Language models is what everyone is talking about in the AI industry. We decided to leverage LLMs and create an application that students around the world could use, especially those that might not have access to a professor and want to efficiently learn complex topics while using their favorite STEM YouTube playlists and/or notes, EffiSTEM does just that and more while offering a customizable learning experience for each user based on their learning preferences and even hobbies, using analogies and a language that a given user can better understand. Learning concepts in STEM has never been this easy.

What it does

EffiSTEM allows users to specify their preferred way of learning material which allows the large language model to approach explanations in a way that benefits the student. We discovered that if a user specifies for their hobbies for instance, the LLM will use that to it's advantage and explain a technical concept using a vocabulary and analogies that the given student understands. The user specifies a YouTube playlist and also has the choice of uploading notes for a STEM subject which allows the Large Language Model to have access to relevant material that interests the student, i.e., the student is able to talk with the content embedded within the YouTube playlist and notes uploaded. EffiSTEM allows you to study better, whether you have an exam the upcoming day or want to learn without an instructor, EffiSTEM is here to assist you at any time anywhere.

How we built it

Our team leveraged React.js in the Front-End, and Flask in the Back-End. We used multiple API's such as Googles Speech to Text API, MathPix for SOTA STEM OCR, and LangChain for optimal LLM development and deployment. Instruction fine-tuning was used as a way to reduce model hallucinations (Temperature is also set to 0).

Challenges we ran into

Organizing a team of only 3 individuals to complete a very large projects in such a short time-frame. Finding out creative ways to use LLMs

Accomplishments that we're proud of

As a team we were able to learn and develop a unique application that students around the world would use in a heartbeat. EffiSTEM has the potential to reduce the amount of time student have to study, and greatly improves the experience a student has when interacting with chatbots from a STEM standpoint.

What we learned

We learned how to leverage LLMs for creating a customized learning experience for the user. As a team we are much more comfortable in working on large projects and implementing functionalities from a Full-Stack Development standpoint.

What's next for EffiSTEM

The next steps for EffiSTEM is to add the capabilities to receive feedback from the user and therefore adapt the outputs based on the given feedback (RLHF). Our team also wants to look into efficiently fine tune LLMs such that they are better adapted for our given use case and further instruction fine-tune our LLMs in order to reduce the chances of hallucinating.

Built With

Share this project:

Updates