Index Won/Lost Call Transcripts into a Searchable Objection Playbook

This is the ingestion half of an objection-handling RAG system. Every time a new call transcript lands in Dropbox, it extracts the moments where a prospect raised an objection and the rep responded, tags each pair with the deal outcome (won or lost), generates embeddings, and stores the indexed snippets in a Notion database. The result is a clean, queryable corpus that downstream retrieval workflows draw from.

Run this once to stand up the knowledge base, then leave it on. Use it when your team records calls (Gong, Zoom, Fireflies) and exports transcripts to Dropbox, and you want every conversation to feed a living playbook instead of dying in a folder.

1. 1A new file in the watched Dropbox folder triggers the flow.
2. 2The transcript text is downloaded and read.
3. 3OpenAI segments the call into objection -> response pairs and labels the deal outcome and topic.
4. 4Each pair is embedded for semantic search.
5. 5A filter drops fragments shorter than a usable threshold so noise never enters the index.
6. 6The clean, embedded snippets are upserted into the Notion playbook database, one row per objection with a link back to the source transcript.