Answers to any question about your favorite podcast episodes.
Podquest searches and retrieves podcast episodes from the open source podcast index.
It then grabs those mp3 urls, downloads them, and transcribes them with the open-source whisper API.
With that transcript, it then creates embeddings which it stores in a vector database.
A user can then query the textual content of the selected podcast episode using an AI large language model. ChatGPT is the LLM model currently, though I will be looking to replace that with a responsible open source alternative.
Podquest checks the user query against the vector database using FAISS to find the closest "answer" to the users query.
Langchain is used to orchestrate all of the various pieces.
There are 2 modes in the experience:
- Questions mode: A simple query mode that answers questions about what was said in the podcast.
- Jump mode: Implements a podcast player, and allows you to ask questions about when something was discussed in the podcast, and skip to the point where it was mentioned.
I would love contributions, questions, feedback, anything. Please feel free to open an issue or PR.