My favorite podcast is the All-In Pod, but search and discovery is tough for podcasts.

So I built a semantic search index that lets you search across every episode of the pod with Google-level accuracy.

It uses OpenAI's latest ada-2 embedding model and Pinecone to index the embeddings + k-NN search. Wrapped up in a Next.js + Vercel webapp.

The project is designed to work with any YouTube channel / playlist. I'm just using All-In as a demo.

Would love to hear feedback & suggestions.

Very cool, how are you doing the speech-to-text part, with Whisper?

The about page explains it. Used a "hacky" method in retrieving the YT transcripts.

Great job! Love this. What other improvements besides using whisper did you have in mind?

Honestly I haven't thought too much about it. Just having fun building :)