Create a chatbot that can answer any question based on a podcast episode

[kavaivaleri]

Let visitors ask natural-language questions about a specific podcast episode (or the entire library) and get concise, timestamped answers with links to listen/watch that exact moment.

Related issues: #61, #62, #64

[gupta-vivek]

Do you have any preference for which vector database, LLM, and framework should be used?

[alexeygrigorev]

I'd go with minsearch without vectors because it's a lot easier to maintain

As for the framework, plain openai API, agents SDK or Pydantic AI. You can also try toyaikit

[alexeygrigorev]

I'd also use groq as they have quite generous free tier

[alexeygrigorev]

I actually don't think we'll need any agents library, if we just implement simple rag with groq and minsearch it'll be already good enough

[gupta-vivek]

Yeah, if it's a simple rag, additional framework won't be required. Will need that only if additional features/functionalities need to be implemented. I shall go ahead with minsearch and Groq.