College Track
Inspiration
When COVID 19 hit, online learning tools like Youtube became more prevalent and accessible. With lectures and lessons being posted online, it fed into what Gen Z's like John enjoyed and learned better with. Using Youtube for learning offered agency that students have for controlling the instructional delivery by pausing, rewinding, or forwarding so that they can work at their own pace and focus specifically on where they need support.
Especially in the sciences when students took hands-on lab courses via Youtube videos, studies show that 91% of students report that teachers should use online science instructional videos with their students when classes are taught in-person. Across other disciplines, the majority of students would like to see videos continue to be included as legitimate resources for science learning ecosystems post-pandemic.
This shifting preference is now driving curricula and technological changes in some schools to support increased demands on teachers leveraging Youtube to support curriculum-based instruction.
What it does
That's why we created Summarizr. Summarizr leverages AssemblyAI's API to divide Youtube video into chapters so that students can head right ot the topics that interest them. Instead of having to rewind to catch what's narrated or said in the video, videos are also transcribed so that students can easily copy and paste what was said. Of course, our app also summarizes and extracts main points in the video and tells users the timestamp durations that the summary is for. This is especially useful when students need to cover a lot of ground fast.
Additionally, Summarizr has a trained machine learning model from Tensorflow to help answer questions about facts drawn from content in the Youtube video. So if a video covers how recommendation systems work, you can ask a question about how markov chains work, and our app would excerpt a relevant section of the video and summarize how markov chains were explained in the video.
How It's Built
We built this project using React, Tailwind CSS, and Material UI for the front end, and flask and python for the back end. The Question and Answer machine learning model is powered by Tensorflow's BERT, which Pratyay will explain more in the next slide.
Challenges We Faced
With half the team working on the front end who are novice at React, there was a quick learning curve we had to overcome. And with many projects, connecting the front end to the back end is always an adventure.
What We're Proud Of
We trained a machine learning model to answer questions in the span of a weekend! We also became more familiar with a new language.
Log in or sign up for Devpost to join the conversation.