Which episode should I listen to?
Tell the concierge a mood, a feeling, or a question — it reads all 55 conversations and makes the call.
Notes on methodology
Recommendations come from a single LLM call (via OpenRouter): your message plus a compact catalog of every episode (summary, dominant topics, guest, length). Every pick is validated against the real episode list server-side — the model can’t invent one — and it explains its reasoning.
Your text is sent to the recommendation endpoint and on to the model provider to generate the answer; nothing is stored beyond a short-lived cache of popular queries.