How saarthakkj/detoxify_yt works
Main Components
A. Client-Side Components:
• Chrome Extension (Frontend)
• Popup Interface
• Content Script
• Background Service Worker
B. Server-Side Components:
• FastAPI Backend
• BERT Classification Model
• BrightData API Integration
Specific Implementation Details
a) Extension Components:
• Popup interface for category selection
• Content script for YouTube page interaction
• Background script for API communication
b) Backend Flow:
• FastAPI endpoints
• Token-based authentication
• Asynchronous processing
• Model inference pipeline
c) Model Integration:
• BERT classification system
• Multi-category output (Chess, Coding, Mathematics)
• Real-time inference path
Key Processes to Illustrate
Main workflows:
Additional Elements
Include:
• Performance metrics (87.8% accuracy)
• Category labels
• Security components
• API authentication flow
Technical Details to Note
Highlight:
• Python 3.8+ requirement
• FastAPI/Uvicorn backend
• BERT model specifications
• Chrome extension manifest version
Remember to include the performance metrics (ROC-AUC scores) for each category as they're significant differentiators:
• Chess: 0.976
• Coding: 0.971
• Mathematics: 0.949
• Other: 0.941
This system design should emphasize the project's unique combination of browser extension technology with AI-powered content classification, making it clear how the components work together to provide a "cleaner YouTube experience."