I made a similar thing for the Lex Friedman podcast. It transcribes the audio and runs it through a summarization model. You can then search the entire corpus to find relevant snippets from episodes:

http://lexicon.lol