As a student, I have experienced the challenges of self-study and the lack of external validation for the material I learn on my own. Many of my peers have a different approach to learning, relying heavily on instructors and structured learning environments. This led us to recognize the need for a solution that would bridge the gap between tailored learning and self-learning. Our project aims to provide a platform that makes self-study more accessible and efficient, providing personalized learning experiences and efficient methods of testing knowledge. Our goal is to empower learners to take control of their own education and to make self-study a viable and effective option for anyone.
What it does
Our project is the first tool where people can create self-assessments tailored only to the required material. This allows learners to focus on specific topics or areas of study and to test their knowledge in a way that is tailored to the material being studied.
Our project would benefit learners, educators, and anyone interested in self-study or self-education. It provides an easy and effective way to self-assess and achieve an active learning dynamic on their own, allowing learners to learn at their own pace and in a way that is tailored to their individual needs and interests. For educators, our platform saves time and effort by providing a tool for creating question banks and valid options for assessment, allowing them to focus more on teaching and less on administrative tasks.
How we built it
We started by brainstorming ideas for how to best extract text from media files and decided to use the assemblyai API to convert audio and video into text. The app also offers quizzes on any topic of your choice and this is simply done via a prompt.
Next, we used state-of-the-art models from the cohere library to summarize the text content and extract key information. We then used OpenAI's GPT-3 language model to generate questions, options, and explanations for the quizzes.
To ensure that our solution was able to handle multiple languages, we integrated the Google Translate API to detect the input language and allow users to take the quiz in their choice of language.
Challenges we ran into
|Inconsistent output format from GPT-3 making it hard to parse questions, options, and explanations||Set the model temperature to 0.0 for a more consistent output and use a robust regex expression|
|Slow processing||Making parallel calls where possible to reduce request time|
|Some texts are hard to extract multiple choice options from due to complexity||Turn these questions into long answer questions|
|Identifying the correct option from GPT-3's generated explanations||More prompt engineering (making the first option always the right answer, randomizing happens in the front end)|
What's next for Material Digest
In addition to providing a platform for creating personalized question banks and testing knowledge, our project has several other features that would benefit learners and educators. For example,
- A live streaming access for convenience, just submit a link to a live audio/video recording and get questions generated automatically at the end of the live session
- Additionally, our platform would include analytics and reporting tools to track progress and identify areas for improvement.
- Quiz export, users would be able to send quizzes to PDFs for offline learning/printing. Google forms also.
- Adaptive learning. The more questions you get, the harder the quizzes would get. -Question answering, users can create quizzes and we will try to answer them given the provided context.
In the future, we plan to continue refining and improving our solution to make self-education even more accessible and effective.