ScrollSense

Inspiration

We noticed that most people spend hours on YouTube and other platforms without realizing how much of the content is misleading, sensationalized, or unproductive. Studies suggest that up to 60% of online content can be unverified or fake, and it’s easy to fall into endless “rabbit holes.” We wanted a tool to help users track and understand their media consumption, giving insights into what’s credible versus what’s just clickbait or entertainment.

What it does

ScrollSense tracks your video consumption in real time, sending video metadata (title, channel, URL) to a backend, and analyzes it to show a daily “media diet”. Users can see:

How much time they spent on credible content
How much time was unverified or misleading
How much was purely entertainment

The tool prepares the foundation for AI-based credibility scoring and visualizes consumption trends in an easy-to-read dashboard.

How we built it

Frontend: Chrome extension (content script + service worker), React/Next.js dashboard
Backend: Python with FastAPI
Data processing: Pandas, NumPy
AI classification (planned): Gemini AI for credibility scoring

Videos are detected in real time, sent to the backend, and stored as JSON objects. The dashboard updates dynamically to reflect daily statistics.

Challenges we ran into

Single-page navigation on YouTube required tracking videoId because pages don’t fully reload.
CORS errors when connecting the Chrome extension to a local FastAPI backend.
Managing multiple Python virtual environments and dependency issues on macOS.
Handling async fetch calls and ensuring the dashboard updates in real time.

Accomplishments that we're proud of

Real-time detection of videos across YouTube’s SPA interface.
Fully working FastAPI backend storing video data.
Dashboard that updates dynamically and visualizes users’ “media diet.”
Successfully bridging a Chrome extension with a Python backend for a real-time demo.

What we learned

How to build and debug Chrome extensions with content scripts and service workers.
Connecting frontend extensions to a backend API securely.
Handling async behavior in single-page apps.
Modeling media consumption data for future AI classification.

What's next for ScrollSense

Integrate Gemini AI to classify videos as credible, fake, or entertainment.
Add daily/weekly summaries and personalized insights.
Expand to other platforms beyond YouTube.
Provide browser notifications when suspicious content is detected in real time.