Turn Notes into RAG Study Flashcards with n8n, OpenAI, and Slack

Study notes in → chunked, embedded, and retrieved → LLM-generated Q&A cards → human review → Slack out

Who

Learners or knowledge workers who need to quickly turn large amounts of study notes into reviewable flashcards

You get

Notes are chunked, vector-retrieved, and LLM-generated into a reviewed flashcard deck with SM-2 spaced-repetition parameters

You need

n8n · slack

Steps

1. Configure credentials and base parameters

   Add OpenAI credentials (for embeddings and chat completion) and Slack API credentials (chat:write required) in n8n. Open the Set - Config node and verify fields: embeddingApiUrl, embeddingModel, chunkSizeChars, chunkOverlapChars, retrievalTopK, similarityThreshold, flashcardCount, defaultTopic, notifyChannel. If using Webhook trigger, copy the webhook URL and configure your external system to POST notesText to that endpoint.

2. Trigger, chunk notes, and wait for topic input

   Receive notes via Webhook - Notes Submitted, or use Manual Trigger - Test Run for testing. The JS - Chunk Notes & Prepare Embedding Requests node recursively splits notes into paragraph/sentence-aware chunks with a trailing overlap window and markdown heading metadata, outputting a chunks array. Then Wait - For Topic/Scope Selection pauses execution until you specify the topic or scope for flashcard generation.

3. Batch embeddings, MMR retrieval, and card generation

   The HTTP Request - Get Embeddings node sends a single batched request to OpenAI for embeddings of both the topic query and all chunks. JS - Vector Similarity Search & Context Assembly computes cosine similarity, filters by threshold, then runs a Maximal Marginal Relevance (MMR) loop to select relevant yet non-redundant top-K chunks, assembling them into retrieved context. AI - Generate Flashcards from Context calls the OpenAI chat model to generate JSON-format Q&A cards strictly grounded in that retrieved context.

   Execution prompt · Source text。AI - Generate Flashcards from Context node text parameter (n8n LLM Chain prompt field)

   Copy prompt

   =You are a study-flashcard writer. Using ONLY the retrieved context below (do not invent facts not present in it), write {{ $('JS - Vector Similarity Search & Context Assembly').item.json.flashcardCount }} high-quality flashcards on the topic: "{{ $('JS - Vector Similarity Search & Context Assembly').item.json.topic }}".
   
   Retrieved context:
   {{ $('JS - Vector Similarity Search & Context Assembly').item.json.retrievedContext }}
   
   Each card should test one clear, specific fact or concept (avoid vague or compound questions). Vary difficulty across the deck. Return ONLY a JSON array (no markdown fences, no commentary), each element an object with keys: question, answer, tags (array of short keyword strings), difficulty ('easy'|'medium'|'hard'), sourceHeading (the heading of the source passage it came from, or null).

4. Validate, human review, and export

   The JS - Parse, Validate, Dedupe & Assign SM-2 Scheduling node safely parses the LLM's JSON, drops cards missing question/answer, deduplicates near-duplicates using Jaccard similarity over word sets, and initializes SM-2 spaced-repetition state (easinessFactor=2.5, intervalDays=0, repetitions=0, nextReviewDate=today) for each surviving card. Wait - For Flashcard Review & Edits pauses for human review; IF - Review Approved gates delivery. Approved decks proceed to Wait - For Export Confirmation for final confirmation, then IF - Export Approved triggers Slack - Send Flashcard Deck to post to the configured channel; declined decks are marked discarded via Set - Discarded/Cancelled.

   Execution prompt · Source text。Slack - Send Flashcard Deck node text parameter (Slack message content field)

   Copy prompt

   =🗂️ *Flashcard Deck: {{ $json.topic }}*
   
   *Cards:* {{ $json.flashcardCount }} (deduped {{ $json.duplicatesRemoved }}, dropped {{ $json.invalidCardsDropped }} invalid)
   
   {{ $json.flashcards.map((c, i) => (i+1) + '. Q: ' + c.question + '\n   A: ' + c.answer + '  [' + c.difficulty + ']').join('\n\n') }}

You are an n8n workflow setup assistant. The user wants to replicate the "Notes → RAG Flashcards → Slack" template. First confirm they have: 1) n8n instance access, 2) OpenAI API key, 3) Slack bot token (chat:write). Then guide them to import the template from n8n.io/workflows/18210, check defaults in Set - Config, configure credentials per step, and suggest using Manual Trigger for the first end-to-end test.