Turn research papers into teaching materials and exam questions

A research paper lands in your inbox. The content is gold: cutting research, important findings, clear methodology. But it's forty pages of dense academic prose. Your students need this material transformed into something useful: a lesson plan for Monday, a quiz by Wednesday, flashcards for revision. You spend the weekend reading, re-reading, typing notes, organising concepts, writing questions. It's thorough work. It's also work that happens to be repetitive, rule-based, and perfect for automation. Most educators accept this as the cost of doing the job well. The alternative seems to be lower quality materials or, worse, using the paper as-is and watching students struggle through the jargon. But there's a third option. By connecting a PDF dialogue tool, a paraphrasing engine, and a spaced repetition system with a simple orchestration layer, you can transform a research paper into a complete teaching package in minutes rather than hours. This workflow isn't about replacing your judgment or cutting corners. It's about moving past the mechanical parts of the work so you can focus on what requires your actual expertise: selecting sources, setting learning objectives, and reviewing output for accuracy and pedagogy. Everything else can run on schedule.

The Automated Workflow

The core idea is straightforward: extract key concepts and findings from the paper, convert them into student-friendly language, then package them as quiz questions and spaced repetition flashcards. The automation handles extraction and rewriting; you handle validation. We'll use n8n as the orchestration platform. It's self-hosted or cloud-based, integrates cleanly with all three tools, and gives you good visibility into each step. Make would work equally well; Zapier's free tier is too limited for this complexity.

Step 1: Trigger and PDF upload

The workflow starts when you upload a research paper to a designated folder in Google Drive or Dropbox. n8n watches that folder and pulls the file path.

CopyTrigger: Google Drive watch (or Dropbox watch)
Output: file_id, file_name, file_path

Step 2: Extract content from the PDF

Chat With PDF by Copilot.us exposes an API endpoint that accepts a document upload and a query. You'll send the PDF and ask it to extract the paper's main research question, key findings, methodology overview, and conclusions.

CopyPOST https://api.copilot.us/v1/pdf/query
{ "file_path": "{{ $node.trigger.data.file_path }}", "query": "Summarise this paper in three parts: 1) Research question and hypothesis, 2) Key findings and data, 3) Conclusions and implications. Be concise."
}

n8n will receive back structured text. Store this in a variable for the next steps.

CopyOutput: extracted_summary (string)

Step 3: Paraphrase for student comprehension

QuillBot's API accepts text and returns paraphrased versions. You'll send the extracted summary and ask for a simplified, student-friendly version. This step is crucial because academic summaries still carry jargon; paraphrasing lowers the reading level without losing meaning.

CopyPOST https://api.quillbot.com/batch
{ "texts": ["{{ $node.step2.data.extracted_summary }}"], "mode": "standard", "intensity": "moderate", "tone": "formal"
}

Set the intensity to moderate; high intensity risks oversimplifying. Store the paraphrased text.

CopyOutput: student_friendly_summary (string)

Step 4: Generate quiz questions

Now use an LLM (Claude Opus 4.6 via API) to turn the paraphrased summary into five multiple-choice quiz questions and five short-answer questions. Claude is excellent at educational content.

CopyPOST https://api.anthropic.com/v1/messages
{ "model": "claude-opus-4.6", "max_tokens": 2000, "messages": [ { "role": "user", "content": "Based on the following summary, create 5 multiple-choice questions (with four options each, labelled A-D) and 5 short-answer questions suitable for university students. Use the summary only; do not add outside knowledge.\n\nSummary:\n{{ $node.step3.data.student_friendly_summary }}" } ]
}

Parse the response to separate multiple-choice from short-answer questions. Store both.

CopyOutput: mc_questions (array), short_answer_questions (array)

Step 5: Create flashcard data

Send the paraphrased summary to Claude again with a different prompt: extract ten key concepts from the paper and create flashcard pairs (concept / definition or explanation).

CopyPOST https://api.anthropic.com/v1/messages
{ "model": "claude-opus-4.6", "max_tokens": 1500, "messages": [ { "role": "user", "content": "Extract 10 key concepts or terms from this summary. For each, create a flashcard with a front side (the concept or question) and a back side (definition, explanation, or answer). Format as JSON array with objects containing 'front' and 'back' keys.\n\nSummary:\n{{ $node.step3.data.student_friendly_summary }}" } ]
}

Claude should return valid JSON. Validate and store it.

CopyOutput: flashcards (array of objects)

Step 6: Push quiz questions to a database or document

Create a Google Doc or Airtable record containing the quiz questions and answers. This becomes a reusable resource you can distribute or edit later.

CopyPOST https://www.googleapis.com/drive/v3/files
{ "name": "{{ $node.trigger.data.file_name }} - Quiz Questions", "mimeType": "application/vnd.google-apps.document", "parents": ["{{ quiz_folder_id }}"]
}

Then insert the content (questions and answer keys) into the document using the Google Docs API.

Step 7: Sync flashcards to Rember

Rember has a REST API for importing flashcard sets. You'll POST the flashcard array directly.

CopyPOST https://api.rember.ai/v1/decks
{ "name": "{{ $node.trigger.data.file_name }} - Key Concepts", "cards": {{ $node.step5.data.flashcards }}
}

This creates a new deck in your Rember account, ready for students to start using spaced repetition.

Step 8: Send you a summary

Finally, send yourself (via email or Slack) a notification that includes the paper name, the number of questions created, and links to the quiz document and flashcard deck. This confirms everything worked and lets you quickly review the output.

CopyPOST https://hooks.slack.com/services/YOUR/WEBHOOK/URL
{ "text": "Research paper processed: {{ $node.trigger.data.file_name }}\nQuiz: {{ $node.step6.data.doc_link }}\nFlashcards: {{ $node.step7.data.deck_link }}"
}

The entire flow, once set up, takes about two minutes to complete for a typical ten-page paper. You spend that time reviewing the output, not generating it.

The Manual Alternative

If you prefer not to automate, the process is straightforward but time-consuming. Upload your PDF to Chat With PDF, ask it to summarise the paper in sections. Copy that summary into QuillBot and paraphrase it. Open a text editor and manually write quiz questions based on the paraphrased text. Finally, log into Rember and hand-type flashcards. This method gives you total control at every step, which is valuable if your papers are highly specialised or if you want to customise quiz difficulty to match your specific cohort. The trade-off is obvious: it takes forty-five minutes to an hour per paper instead of five minutes. Do this for five papers a term, and you've recovered a full working day.

Pro Tips

Error handling and PDF failures

Chat With PDF sometimes struggles with scanned PDFs or documents with unusual formatting.

Before automating a batch, test the API with your PDF first. If extraction fails, the workflow should pause and send you an alert rather than passing garbage downstream. Add a conditional step in n8n that checks whether extracted_summary contains meaningful content (word count > 100, no error messages) before proceeding.

Rate limits and costs

The Copilot.us API has rate limits; check documentation for current limits, but assume 100 requests per minute if using a paid tier. If you're processing multiple papers in a batch, stagger the triggers or queue them. Claude API has separate rate limits; the Opus 4.6 model is more expensive per token but produces better educational content than cheaper alternatives. Budget for roughly 2000 tokens per paper.

Paraphrasing intensity and tone

QuillBot's "moderate" intensity usually works well for academic content. If students still report difficulty, bump it to "high" on a second pass, but review output carefully; extreme paraphrasing can introduce inaccuracies. Avoid the "creative" tone; stick with "formal" for academic papers.

Validating question quality

LLMs occasionally generate plausible-sounding but incorrect quiz questions, especially for highly technical papers. Before distributing quizzes to students, spot-check questions against the original paper. If you find errors, update your Claude prompt to be more specific about accuracy.

Rember deck sharing

Once a flashcard deck is created in Rember, you can share a link with students or export the deck as JSON for use in other spaced repetition apps. Rember's own interface is clean, but some students prefer Anki or RemNote. Keep the JSON export handy so you're not locked in.

Cost Breakdown