I watch a lot of YouTube. Too much, probably. Specifically tech channels — AI researchers, startup founders, security researchers, developers explaining new tools. The problem isn't finding good content. It's that video is a terrible format for staying current. A 45-minute interview might contain 3 minutes of genuinely new insight buried under intros, sponsors, and tangents.
So I started building BuzzRAG.
The Idea
Take YouTube videos I'm already subscribed to. Transcribe them. Have AI extract the signal from the noise and write it up as a readable article. Basically: my YouTube subscriptions, but as a news site.
The first commit went in on October 21st. Next.js, Prisma, Google OAuth to pull my YouTube subscriptions. Deployed to Vercel.
Day One Reality
The first night was a wrestling match with deployment. 14 commits just trying to get Prisma working on Vercel's serverless runtime. Binary targets, engine paths, tracing configs. The kind of yak-shaving that makes you question your life choices at 10pm.
But by the end of it, I had a working feed. Log in with Google, see your YouTube subscriptions' latest videos. Not impressive, but real.
Week One: Making It Useful
Over the next few days, things came together fast:
- Watch queue — bookmark videos to watch later
- Shorts detection — filter out YouTube Shorts (they're not useful for this)
- PWA support — installed it on my phone like a native app
- Search — filter the feed by channel name or video title
- Infinite scroll — cursor-based pagination instead of the janky page-based approach
The feed was genuinely useful at this point. I was opening it instead of YouTube to see what was new. That felt like validation.
What's Next
The feed is just step one. The real value is in the transcription and AI summarization pipeline. That's what turns "another YouTube video" into "I can read this in 2 minutes."
Starting on that next.