Upsert Notion Docs to Supabase Vector Store for RAG

This workflow enables building a Retrieval-Augmented Generation (RAG) system on dynamic Notion knowledge base data. It polls a Notion database every minute for updates, processes changed pages sequentially, deletes existing embeddings via page ID metadata, retrieves all blocks, combines them into a single string, generates embeddings, splits into chunks if needed, and upserts with metadata into Supabase vector store. Adaptable to PGVector, Pinecone, or Qdrant.

Key benefits include maintaining fresh embeddings without data loss, improved accuracy over cached polling, and seamless integration for AI-powered search and Q&A. The built-in question-answering chain allows instant querying of embedded content via chat.

Ideal use cases: Powering chatbots with live Notion docs for customer support, internal knowledge bases in marketing/event management, or research tools. Saves hours of manual data syncing and enables scalable AI applications on structured Notion content.