Inspiration
A popular distaste for long, uninformative videos and the removal of the dislike button as a means to determine the quality of a YouTube video's content.
What it does
Given a YouTube video URL, the site will retrieve a transcript of the video from YouTube, summarize it through openAI's API, extract named entities from the transcript and supply both a definition and a generated YouTube url-embedded timestamp for when each entity is mentioned in the video. This allows a person to effectively ‘parse’ a YouTube video almost instantly to generate an efficient text summary and url-embedded time stamps to when important subjects are talked about in the video with accompanying Wikipedia definitions.
How we built it
Caijun and Gabriel handled various backend functionalities, Matthew handled building the Flask app and connect the front and backend and Patrick handled functions involving OpenAI
Challenges we ran into
Attempting to host the site on a local server, employing clunky UI builders, misunderstanding the functionality of various API's and debugging a remotely hosted application
Accomplishments that we're proud of
The entire thing
What we learned
How to host sites on Linode, general python skill buff, OpenAI API usage, spaCy language model usage, HTML, CSS and perseverance.
What's next for Sherlock
Our next steps are to turn the site into a more useful chrome extension that can display summaries and insights while you browse YouTube. And implement more powerful keyword extraction and the ability to search through keywords.
Log in or sign up for Devpost to join the conversation.