VideoWatch
AI-powered video monitoring platform that scrapes and tracks videos across multiple sites.
VideoWatch started as a personal need to monitor video content across multiple platforms without manually checking each site. The challenge was building a scraper that could handle JavaScript-heavy sites, bypass bot detection, and scale without consuming too much server resources.
The solution uses Python FastAPI for the backend, Playwright for headless browser scraping, and yt-dlp for platform-native extraction (YouTube, VK Video). A scheduler runs automatic scans and an SSE-powered live feed shows real-time progress. The app is deployed on Oracle Cloud Always Free tier with GitHub webhook auto-deployment.
Key engineering challenges included optimizing CPU usage on a 1 OCPU server, fixing VK Video's duplicate title/thumbnail bug caused by DOM shadowing in the JS extraction, and building a robust deduplication system that handles URL token rotation without breaking links.