RAG Study Flashcards Workflow: OpenAI + Slack in n8n

Turn your study notes into reviewed flashcards—with OpenAI-powered RAG and Slack delivery

Bring your study materials to this n8n workflow and get a grounded flashcard deck generated using Retrieval-Augmented Generation (RAG). The workflow chunks your notes, retrieves the most relevant passages with OpenAI embeddings, generates JSON flashcards strictly from that context, and sends the approved deck to a Slack channel for easy review and use.

What this workflow does

• Ingest notes via a webhook: Accepts study notes through a POST webhook (or runs manually for testing), then loads configuration for chunking, retrieval, flashcard count, and Slack delivery.
• Chunk and scope your content: Splits the notes into overlapping, heading-aware text chunks and pauses to collect the topic/scope you want flashcards for.
• Retrieve relevant passages (RAG): Requests OpenAI embeddings in a batched call for the topic query and all note chunks, ranks chunks by cosine similarity, applies diversity-aware selection (MMR), and assembles the retrieved context.
• Generate grounded flashcards: Uses an OpenAI chat model to produce a JSON array of flashcards strictly grounded in the retrieved context.
• Validate, deduplicate, and schedule: Validates and deduplicates generated cards, initializes SM-2 scheduling fields, then pauses for human review and final export approval.
• Publish to Slack: Sends the approved flashcard deck to your configured Slack channel, or marks the run as discarded if review/export is declined.

Use cases

• Convert lecture notes or reading notes into a flashcard deck for spaced repetition using SM-2.
• Create topic-specific flashcards (e.g., “Key concepts in X”) by selecting a scope during the workflow pause step.
• Automate study material review for teams—approve once, share to Slack for consistent learning assets.

Technical details

• Integrations: OpenAI (embeddings + chat completion) and Slack
• n8n nodes/logic used: if, set, code, wait, slack, webhook
• Core workflow behavior: batched embedding calls, cosine similarity ranking, MMR diversity selection, JSON generation, validation/deduplication, SM-2 initialization, and human approval gating