SECOND DRAFT: PODCAST TRANSCRIPT SEARCH STARTER KIT

Original fictional interview excerpts for the Quickchat AI tutorial:
https://quickchat.ai/post/podcast-transcript-search-assistant

These are original teaching examples, not recordings, complete podcast episodes, Riverside exports, customer data or a publicly deployed podcast service. You may reuse these excerpts to follow the tutorial. Their time labels belong to the example text; there is no audio player to seek.

FILES
- episode-18.txt: Nina Park on creative routines; two source timecodes.
- episode-27.txt: Mara Bell on interviewing; two source timecodes.
- episode-34.txt: Sam Rivera on editing and room tone; no source timecodes.
- episode-inventory.csv: source identity, coverage and rights inventory.
- transcript-header-template.txt: a header for your own authorized exports.
- main-prompt.txt: the tested Agent instructions.
- source-checks.csv: eleven manual checks, including two conversational follow-ups.
- dataset-messages.txt: eight standalone messages ready for Testing's Paste messages field.
- evaluation-criteria.txt: criteria plus full literal source passages for checking quotations.

FOLLOW THE TUTORIAL
1. For your own show, export a transcript from the final edited recording in Riverside, Descript or your publishing tool. Keep real speaker names and source timecodes. These example files let you learn without exporting anything first.
2. Create a separate empty Quickchat AI Agent named Second Draft Search.
3. Add each episode text as a separate Knowledge Base article. Use its episode number/title as the title and its Episode URL in the URL field. Preserve the full text and wait for Ready. Upload File is also available; inspect the extracted text afterward.
4. Open Identity > AI Persona > Main Prompt > Edit and paste main-prompt.txt. The change saves automatically.
5. Try the control question, then the remembered-moment question in a fresh conversation. Follow its episode link and check the passage. The sample source pages become publicly available when the tutorial is published.
6. Follow source-checks.csv. The followUp row must follow opening in the same conversation; narrowedPauses must follow broadPauses. Every other row starts fresh.
7. In Conversations > Testing > Create Dataset, paste dataset-messages.txt into Paste messages. Each line containing only --- separates two standalone questions. Paste evaluation-criteria.txt into Evaluation criteria. Run the dataset and inspect answers as well as scores. AI replies and testing consume account credits.
8. When ready for your audience, open Channels > Your Website > Install and choose a supported installation. Configure Availability and access, Appearance and Text Content. Verify the visitor experience on desktop and a phone.

This ZIP is a set of editable learning files, not a one-click Agent import or a CSV schema guaranteed for every import tool. source-checks.csv is a manual worksheet; use dataset-messages.txt for the demonstrated Testing workflow.

For a real archive, replace the sample show/guest names, URLs and excerpt assumptions in the files and prompt. Transcribe audio separately. Never invent missing guests, quotations, timecodes or playback URLs. A timestamp is only as accurate as its alignment with the final published audio.
