Replicating YouTube Shorts recommendations with LLMs and semantic embeddings

amyxzh · x · 2026-08-20

Implemented a personalized recommendation experience on YouTube Shorts using a browser extension that intercepts upcoming content, interprets video metadata via LLMs, and matches them to natural language preferences using semantic embeddings. The system also includes embedded micro-feedback mechanisms and DOM injections for immediate content changes to address algorithmic latency.

Related event: Researchers Build LLM-Powered Continuous Alignment for YouTube Shorts(4 posts)→

Original post →

More from coding & agent

coding & agent channel →