Turn a PDF from Google Drive into a chat ready knowledge base and answer questions through a simple webhook. Great for customer support teams that need fast, accurate answers from trusted documents.

The flow ingests a PDF from Google Drive, splits the text into readable chunks, and creates embeddings with OpenAI. It saves the vectors in a Qdrant collection. When a POST request hits the webhook with an input and a company value, the system selects the matching Qdrant collection, retrieves the top 5 relevant chunks, and uses an OpenAI chat model to form a clear answer. The result is sent back instantly through the response node. You can also test live with the manual chat trigger in n8n.

Setup needs accounts for Google Drive, OpenAI, and Qdrant. Provide a Google Drive file ID, an OpenAI API key, and a Qdrant URL and key. Send POST requests that include input and company fields to route queries to the right collection. Expect shorter handle time, fewer repeated questions for agents, and the ability to scale support without adding headcount. Useful for help centers, onboarding guides, and policy FAQ support.

• Downloads a specific PDF from Google Drive using a file ID
• Converts the PDF binary into text documents for processing
• Splits text into chunks of 3000 characters with 200 character overlap
• Generates vector embeddings with OpenAI for high quality search
• Stores vectors in Qdrant and supports dynamic collection names by company
• Retrieves the top 5 most relevant chunks to ground each answer
• Uses an OpenAI chat model to produce clear, natural language responses
• Receives questions via a POST webhook and returns answers instantly
• Includes a manual chat trigger in n8n for quick testing and demos

• Reduce manual research from 30 minutes to under 30 seconds
• Automate up to 80 percent of common customer questions
• Improve answer consistency by using the source PDF as context
• Handle up to 10 times more queries with the same team
• Connect Google Drive, Qdrant, and OpenAI in one flow
• Return answers as API responses for sites and chat widgets

1. Import the template into n8n: Create a new workflow in n8n > Click the three dots menu > Select 'Import from File' > Choose the downloaded JSON file.
2. You'll need accounts with Google Drive, OpenAI, Qdrant and Webhook.site. See the Tools Required section above for links to create accounts with these services.
3. Open the Google Drive node. Double click the node, then in the Credential to connect with dropdown, click Create new credential and follow the on screen steps to connect your Google Drive account. Paste your PDF file ID into the File ID field.
4. Open the OpenAI nodes (Embeddings and Chat). In each node, create a new OpenAI credential and paste your API key from the OpenAI account API keys page. Use a clear name like OpenAI Main to keep it organized.
5. Open both Qdrant nodes. Create a new Qdrant credential with your server URL and API key. Use a descriptive name like Qdrant Prod. Confirm the collection value for the insert node is set to your target collection.
6. Check the text splitter node. Keep chunk size at 3000 and overlap at 200 for balanced context and cost. Adjust only if your documents are very short or very long.
7. Run the ingest path: Click Execute Workflow, then let the Google Drive, splitter, embeddings, and Qdrant insert nodes run. Confirm vectors appear in your Qdrant collection.
8. Open the Webhook node. Copy the Production URL. Set your client or test tool to send a POST with JSON that includes input and company fields. Example: {"input":"What is the refund policy?","company":"crowd"}.
9. Ensure the Retriever is linked to the Qdrant Vector Store with the dynamic collection. Verify topK is set to 5 for relevant context.
10. Send a test request from Webhook.site or curl. Check that the Respond to Webhook node returns a clear text answer. If you see errors, verify OpenAI keys, Qdrant URL and collection names, and the Google Drive file ID.
11. If answers are off topic, re run the ingest after updating the PDF, or tune chunk size and overlap. Make sure the company value matches an existing Qdrant collection with vectors.