Hey everyone! 👋 I built UniFetch because I was constantly jumping between 5-6 different places every time I wanted to download something — checking Telegram channels manually, opening Pirate Bay, Archive.org, hitting dead links, and eventually giving up. There was no single place that brought all these sources together into one clean search. The core idea was simple: type what you want, get links from everywhere in one shot. What started as a basic Telegram scraper using GramJS (MTProto) evolved into something much more interesting. The biggest shift in my thinking came when I realized I was approaching channel discovery completely wrong — I was searching for "GTA Vice City" to find channels, which naturally returned fan clubs and roleplay communities instead of actual download repositories. The correct mental model was like finding a library first, then searching for your book inside it. That realization completely changed the architecture. The system now runs a background cron worker every 6 hours that discovers and validates Telegram channels using a scoring system (checking file ratios, downloadable content types, conversation-to-upload ratios) before ever adding them to the database. Users never search bad channels. Stack is TypeScript/Node.js, Express, MongoDB, Upstash Redis for caching, GramJS for Telegram's MTProto API, and Archive.org's open API as a second source. Zero paid AI APIs in the search path — category detection is pure keyword matching, which honestly performs better than the Gemini-powered version I tried earlier and doesn't hit rate limits. Still early but the foundation is solid. Happy to answer questions about the technical architecture — it was a genuinely interesting problem to build around.