Convert research papers into teaching materials and exam questions

A typical week in your life as an educator involves juggling lesson plans, assessment design, and keeping pace with emerging research in your field. You find a paper that perfectly illustrates the topic you're teaching next week, but converting it into something your students can engage with takes hours. You need to extract the core arguments, identify examples, create discussion prompts, and draft exam questions. By the time you've finished, you've spent an evening you didn't have, and the material still needs polishing. What if that entire process took 30 minutes instead of three hours? The gap between a published paper and a classroom-ready lesson isn't a shortcoming on your part; it's simply where most educators stop using technology. The tools exist to close this gap completely, but they're sitting in separate silos. By connecting them through a simple workflow, you can upload a PDF and receive a structured set of teaching materials within minutes. This workflow uses two purpose-built tools for academic content, plus a lightweight orchestration layer to move data between them without any manual copying or reformatting. The result is a system that takes your research paper and outputs ready-to-use learning objectives, concept summaries, discussion questions, and exam items, all in one go.

The Automated Workflow

The orchestration backbone here is n8n, which is free to self-host and more straightforward than Zapier for this particular job. You could use Make (Integromat) instead if you prefer a managed platform, but n8n's node library gives you direct control over JSON payloads between steps. The workflow sequence is: 1. A PDF lands in a designated folder (Google Drive, Dropbox, or local directory).

2. n8n triggers and extracts the paper metadata and abstract.

3. Chat With PDF by Copilot.us processes the full document and extracts key concepts, research questions, and main findings.

4. Explainpaper highlights technical sections and generates simplified explanations.

5. An LLM (Claude Sonnet 4.6 recommended for education work) synthesises all extracted content into structured teaching materials.

6. The output is formatted and sent to a Google Doc, email, or Airtable base for review.

Setting up the trigger in n8n

Start by installing n8n locally or on a server. Create a new workflow and add a "Google Drive" trigger node. Configure it to watch a specific folder for new PDFs.

CopyTrigger configuration:
- Resource: File
- Event: File Added
- Folder: /Teaching Materials/Papers to Convert
- Mime Type: application/pdf

When a PDF appears, n8n captures the file ID and passes it to the next node.

Extracting metadata with Chat With PDF

Chat With PDF by Copilot.us doesn't have a direct API for workflow integration, but you can work around this by using their web interface via a webhook. Alternatively, use a function node in n8n to call OpenAI's API directly on the PDF content after converting it to base64. If you're using Claude for this step, you'll want the API endpoint for document handling:

CopyPOST https://api.anthropic.com/v1/messages
Content-Type: application/json
Authorization: Bearer YOUR_API_KEY { "model": "claude-sonnet-4.6", "max_tokens": 2000, "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Extract the following from this academic paper: title, authors, abstract, research questions, methodology, main findings, and key concepts. Format as JSON." }, { "type": "document", "source": { "type": "base64", "media_type": "application/pdf", "data": "BASE64_ENCODED_PDF" } } ] } ]
}

Store the response in n8n's internal variables for the next step.

Processing technical sections with Explainpaper

Explainpaper's interface is designed for human interaction, but you can structure your workflow to flag sections that likely contain technical density. Use a subsequent Claude call to identify paragraphs that mention methodology, statistical results, or domain-specific terminology, then prepare those excerpts as a separate output for manual review in Explainpaper if needed. In practice, for most educational workflows, Claude Sonnet 4.6 can simplify technical content directly without requiring Explainpaper as a separate step. However, if your papers use highly specialised notation or domain-specific language, include Explainpaper as a manual checkpoint. The workflow can generate a shareable link with the pre-selected sections highlighted.

Generating teaching materials with Claude

Once you have extracted content, pass everything to Claude Sonnet 4.6 with a detailed prompt that structures the output as JSON. This makes the next step (formatting) trivial.

CopyPOST https://api.anthropic.com/v1/messages
Content-Type: application/json
Authorization: Bearer YOUR_API_KEY { "model": "claude-sonnet-4.6", "max_tokens": 3000, "messages": [ { "role": "user", "content": "Based on the following academic content, generate teaching materials in JSON format with these fields: learning_objectives (array), key_concepts (array of objects with term and explanation), discussion_questions (array with 5-7 questions), exam_questions (array with mix of short-answer and essay prompts). Ensure exam questions vary in difficulty from basic recall to critical analysis.\n\n" + extractedContent } ]
}

The JSON response is structured enough that a function node in n8n can parse it without custom scripting.

Formatting and delivery

Use the final n8n node to write the JSON output to a Google Doc via the Google Docs API. This gives you a human-readable format that you can review, edit, and share immediately.

CopyPOST https://docs.googleapis.com/v1/documents:batchUpdate
Authorization: Bearer GOOGLE_API_TOKEN { "requests": [ { "insertText": { "text": "Learning Objectives:\n" + formattedObjectives + "\n\nKey Concepts:\n" + formattedConcepts + "\n\nDiscussion Questions:\n" + formattedQuestions } } ]
}

Alternatively, send the structured JSON directly to an Airtable base where you maintain a database of teaching materials by subject and term.

The Manual Alternative

If you prefer not to set up n8n, the simplified two-tool approach works well. Upload your PDF to Chat With PDF by Copilot.us and have a conversation with it about the paper's structure and main points. Copy the key information into a document, then paste relevant sections into Explainpaper to clarify any technical language. Finally, open Claude directly (via the web interface or API) and ask it to generate questions and learning objectives based on what you've prepared. This takes longer and requires more copy-pasting, but it gives you fine-grained control at each stage. Many educators prefer this because you can ask follow-up questions and refine outputs in real time.

Pro Tips

Rate limits and batching.

Claude's API has rate limits depending on your plan; if you're processing multiple papers in a day, use n8n's delay node to spread requests across an hour rather than sending them all at once.

This avoids throttling.

PDF quality matters.

Scanned documents with poor OCR will confuse both Chat With PDF and Claude. Test your workflow with a clean, native PDF first. If you often receive scanned papers, add an OCR preprocessing step using a tool like Tesseract or Claude's vision capabilities.

Caching extracted content.

Once you've extracted metadata and key concepts from a paper, store that JSON in Airtable or a spreadsheet. You can reuse it to generate different question sets later, cross-reference papers by theme, or build a searchable library.

Quality gates.

Always review the generated exam questions before sharing with students. LLMs occasionally produce questions with ambiguous wording or answers that aren't cleanly supported by the source material. A 10-minute review saves classroom confusion later.

Cost optimisation.

Use GPT-4.1 mini for the initial extraction step (it's fast and cheap for structured outputs), then reserve Claude Sonnet 4.6 for the final synthesis step where tone and pedagogical reasoning matter more.

Cost Breakdown