Sage AI

Inspiration

While watching YouTube, I often found myself pausing videos to search for answers or context related to what I was watching. This constant context-switching—opening new tabs and searching manually—disrupted the viewing experience. I realized there was no tool that could provide instant, AI-powered answers about the video content directly within YouTube. This gap inspired me

What it does

It goes through the video metadata and transcripts and answers user queries by passing it to an llm.

How we built it

Sage was built as a solo project using JavaScript/JSX, Webpack, and Babel. The extension leverages the OpenRouter API for AI-powered responses and the YouTube Data API for accessing video transcripts and metadata. I focused on creating a seamless sidebar interface that automatically adapts to YouTube's dark and light themes. The build process involved rapid prototyping, frequent testing in Firefox, and iterative UI improvements to ensure a minimal and non-intrusive user experience.

Challenges we ran into

The biggest challenge was aligning and embedding the assistant's sidebar into the YouTube interface in a way that felt native and unobtrusive. YouTube's dynamic layout and frequent UI updates required careful handling to ensure the extension remained stable and visually integrated. Additionally, working with browser extension permissions and content security policies was a new and sometimes tricky aspect of development.

Accomplishments that we're proud of

Got 60+ downloads on Firefox Add-Ons store within 3 days.

What we learned

This project was my first experience building a browser extension. I learned how to work with APIs, manage content scripts, and handle the unique challenges of injecting UI elements into third-party websites. I also deepened my understanding of integrating advanced language models and working with real-time video transcript data.