I have a huge archive of podcasts, which are a great way to get information out of people's heads. However, podcasts are difficult to consume an generate long-term resources from.

I am building a pipeline to automatically reorganize the podcast content into a knowledge base.

1. Use Riverside.fm to record 2. Use Sonix.ai to transform text to speech 3. Use OpenAI to fix grammar and stammer 4. Extract question and answer pairs 5. Use OpenAI to embed questions and answers 6. Use OpenAI with a moving window to extract key questions from the transcript 7. For each key question, use OpenAI with a moving window to extract answers to each key quesiton 8. For each set of key answers, use OpenAI to write an article answering the question

For any given question, including those generated by OpenAI, pull the relevant question and answer pairs from the embedding. Run steps 7 and 8 from above.

For any given quotation, find the transcript from which it came. Find the time index of the question in the transcript. Generate an embedded video clip of the question and answer using either YouTube time index links or Descript or some other tool.

For any given video transcript, check if there are similar quotes from the embedding database. Create a video title linking to a related video if the similarity is very high.

-- SunirShah