NeuroFilterAI

YouTube is the best learning resource on the internet. The problem isn't YouTube. It's that YouTube's recommendation algorithm optimizes for watch time, which is directly opposed to why most people open it to study. NeuroFilter takes a different approach: semantic intent filtering. You state a learning objective — "organic chemistry" or "calculus" or "system design" — and the extension embeds that intent using a local transformer model (Xenova/all-MiniLM-L6-v2 via transformers.js). Every YouTube recommendation gets scored by cosine similarity against that intent. Below-threshold videos get blurred in the DOM. Above-threshold videos stay visible. The key property: the filtering is semantically aware. A video titled "JEE Chemistry | SN2 Reactions" scores high for organic chemistry intent. "Top 10 Gaming Moments 2024" scores near zero. Keyword blockers can't make that distinction — they don't know what you're there to learn. Built for students, developers, researchers, and anyone who uses YouTube intentionally. Technical: MV3 Chrome extension, WASM inference in offscreen document (avoids service worker lifecycle issues), LRU cache for repeat recommendations, ~3s cold start, then near-instant. Solo-built by Piyush, 17, Gorakhpur — during recovery from Ankylosing Spondylitis.