Quick Lecs

Inspiration

Often times, it is difficult to make time for attending lectures due to a myriad of reasons like class projects or interview preparation. Watching recordings is not always the solution, as it may be too time consuming, or the lecture is on a topic you are already familiar with and just want to get a gist of the lecture. This project was made as a way to help students save time, which is arguably our most scarce resource.

What it does

The student is prompted to enter an URL of the video. The video will then be transcribed, and the text will be summarized. The summary is presented to the student for reading.

How we built it

The frontend is built using Flutter. Once the user enters the URL, it is sent to the Flask API. The API is responsible for downloading the video. Then, it converts the video into an audio file. The audio file is uploaded to Google Cloud so that Google Speech-To-Text can transcribe it. Once the text is obtained, we send it to OpenAI's DaVinci model so that it can give us a summary of the text.

Challenges we ran into

We found that the DaVinci model always produced summaries that are only 2 or 3 sentences long. This was not very useful for summarizing lectures as it missed important parts of the video. So, we first broke down the transcripts into chunks and then called the model to obtain a slightly longer and more meaningful summary.

Accomplishments that we're proud of

Since this was our first hackathon, we were unfamiliar with the process. While we have made project spanning a few days in the past, this is our first time making something in the span of 36 hours. It was a learning process to divide tasks such that our work was perfectly optimized.

What we learned

We learned how to use Google AI services like Speech-To-Text. We also learnt about the OpenAI's DaVinci model, and how to fine tune a model to fit our needs. We also learnt about coding on a time crunch.

What's next for Quick Lecs

There are still a few improvement that can be made to QuickLecs to make it even more user friendly. We plan to export the text summary as an audio file for on-the-go listening. We also plan to mark the timestamp of important slides used in the lecture.

Project Leader: Aliya Abdullah Discord: CouchAlu#6724