AI Quiz Auto Grader with RAG, Google Sheets Logging & Slack Alerts

Overview

This workflow automates the grading of quiz submissions using a Retrieval-Augmented Generation (RAG) pipeline. When a student submits a quiz (via webhook), the workflow processes the answer text, stores it as a vector embedding in Pinecone, retrieves relevant context, and uses an AI agent (powered by OpenAI) to grade the response. The grade is then appended to a Google Sheet for record-keeping. If any error occurs during the process, a Slack alert is sent to a designated channel (e.g., #alerts).

This is ideal for educators, training platforms, or any scenario where you need to automatically evaluate open-ended answers against a knowledge base.

Step-by-Step Node Breakdown

1. Webhook Trigger (No auth) — Listens for incoming HTTP POST requests at the path /quiz-auto-grader. The workflow is triggered each time a quiz submission arrives, carrying the student's answer and optional context in the request body.

2. Text Splitter (No auth) — Splits the incoming text into smaller chunks (chunk size: 400 characters, overlap: 40 characters). This prepares the text for embedding and storage, ensuring it fits within the embedding model's token limits.

3. Embeddings (Cohere API Key auth) — Uses Cohere's embed-english-v3.0 model to convert each text chunk into a vector embedding. These embeddings capture semantic meaning for later retrieval.

4. Pinecone Insert (Pinecone API Key auth) — Inserts the generated embeddings (along with the original text) into a Pinecone index named quiz_auto_grader. This builds the vector database that serves as the knowledge base for grading.

5. Pinecone Query (Pinecone API Key auth) — Queries the same Pinecone index to retrieve the most relevant vector results for the incoming quiz answer. This provides contextual information (e.g., correct answers or reference material) that the AI agent can use for grading.

6. Vector Tool (No auth) — Wraps the Pinecone query results into a tool that the AI agent can call. It makes the vector store accessible as a function the agent can invoke to fetch context.

7. Window Memory (No auth) — Maintains a buffer of recent messages (conversation history) to give the AI agent short-term memory. This helps maintain consistency across multiple queries within a single workflow execution.

8. Chat Model (OpenAI API Key auth) — Provides the language model (ChatGPT) that powers the AI agent. No special options are configured; the model uses default parameters.

9. RAG Agent (No auth) — The core reasoning node. It takes the system message "You are an assistant for Quiz Auto Grader" and processes the incoming JSON data ({{ $json }}). It has access to the Vector Tool (for context retrieval), Window Memory (for conversation history), and the Chat Model (for generation). The agent outputs a grade or evaluation.

10. Append Sheet (Google Sheets OAuth2) — Appends the agent's grade (from $json["RAG Agent"].text) to a Google Sheet. The sheet ID and sheet name ("Log") are pre-configured. This logs every graded submission.

11. Slack Alert (Slack OAuth2) — Connected to the error output of the RAG Agent. If any error occurs, it posts a message to the #alerts channel with the error details. This ensures the admin is notified immediately.

Setup Instructions

To use this workflow, you will need:

1. Cohere account — Sign up at Cohere and generate an API key. Add it as a credential in n8n.
2. Pinecone account — Create a Pinecone index named quiz_auto_grader (or change the name in the node). Obtain your API key and environment details, and create a credential in n8n.
3. OpenAI account — Get an API key from OpenAI and add it as a credential.
4. Google Sheets — Create or use an existing spreadsheet. Note the sheet ID (from the URL) and the sheet name (e.g., "Log"). Authorize n8n to access Google Sheets via OAuth2.
5. Slack workspace — Create a Slack app or use an existing bot token that has permission to post messages to a channel. Create a Slack credential in n8n with OAuth2 or a token. Ensure the channel #alerts exists (or change the channel name).
6. Webhook — The workflow exposes a webhook URL (you'll see it after activating the workflow). Configure your quiz submission system to POST JSON data to that URL.

After setting up all credentials, activate the workflow. Each POST to the webhook will trigger the entire pipeline.

Use Cases & Variations

• Course quizzes: Automatically grade open-ended questions based on a stored knowledge base of correct answers and explanations.
• Competency tests: Evaluate candidate responses in recruiting or certification exams.
• Adaptive learning: Modify the grading rubric by changing the system message in the RAG Agent node.
• Alternative vector stores: Replace Pinecone with other supported backends (e.g., Qdrant, Weaviate, Supabase) by swapping the Pinecone nodes.
• Different logging destinations: Instead of Google Sheets, you could log to Airtable, Notion, or a database.
• Multi‑step workflows: Add an email notification node (e.g., SendGrid) to send the grade directly to the student.

{
  "name": "Quiz Auto Grader",
  "nodes": [
    {
      "parameters": {
        "content": "Automated workflow: Quiz Auto Grader",
        "height": 530,
        "width": 1100,
        "color": 5
      },
      "id": "52dc6fbe-f7af-43a8-9226-241e163e6bbf",
      "name": "Sticky Note",
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        -480,
        -240
      ]
    },
    {
      "parameters": {
        "httpMethod": "POST",
        "path": "quiz-auto-grader"
      },
      "id": "f25da8be-15a9-4fe8-b28c-5343b8a88f29",
      "name": "Webhook Trigger",
      "type": "n8n-nodes-base.webhook",
      "typeVersion": 1,
      "position": [
        -300,
        0
      ]
    },
    {
      "parameters": {
        "chunkSize": 400,
        "chunkOverlap": 40
      },
      "id": "62c88abc-3c7d-4fac-9166-2b6d785992f7",
      "name": "Text Splitter",
      "type": "@n8n/n8n-nodes-langchain.textSplitterCharacterTextSplitter",
      "typeVersion": 1,
      "position": [
        -130,
        0
      ]
    },
    {
      "parameters": {
        "model": "embed-english-v3.0",
// ... truncated (copy to see full JSON)