Give fast answers from your documents without manual lookup. The flow turns a Google Drive PDF into a searchable knowledge base and serves answers by API or chat. It helps support teams and internal users get clear replies in seconds.

A manual start downloads the file from Google Drive. The document is split into readable chunks with a recursive splitter set to 3000 characters and 200 overlap. OpenAI embeddings turn each chunk into vectors and the data is stored in a Qdrant collection. Questions come in through an HTTP webhook or a chat trigger. A vector retriever pulls the top five matches, then an OpenAI chat model writes a final answer using a Retrieval QA Chain. The webhook returns clean text to the caller. For multi tenant needs, the retriever can point to a collection named by the company field in the request.

You will need access to the Google Drive file, an OpenAI API key, and a Qdrant endpoint with collections ready. Expect faster response times and less time spent on repeat questions. Good fits include product manuals, policy PDFs, and pricing documents where users ask the same questions again and again.

• Manual trigger downloads a PDF from Google Drive for controlled ingestion
• Document loader uses a recursive text splitter with 3000 size and 200 overlap for better context
• OpenAI embeddings convert text chunks to vectors for semantic search
• Qdrant vector store inserts data into a named collection for fast retrieval
• HTTP webhook accepts POST requests with input and optional company fields
• Chat trigger accepts live chat messages and sends them to the QA chain
• Vector retriever fetches top 5 matches to ground the answer
• OpenAI chat model generates clear answers through a Retrieval QA Chain
• Respond to Webhook returns plain text so any app can consume the result

• Reduce manual answer time from 30 minutes to under 20 seconds
• Handle up to 10 times more questions without extra staff
• Improve answer accuracy by using the best matching document text
• Connect Google Drive, Qdrant, and OpenAI in one flow
• Support multiple clients by routing queries to separate collections

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 and Qdrant. See the Tools Required section above for links to create accounts with these services.
3. In the n8n Credentials manager, open the Google Drive credential or create a new one. Choose Google Drive OAuth and sign in to the Google account that holds the PDF. Save the credential with a clear name.
4. Double click the Google Drive node. Set Operation to Download and paste the file ID of your PDF. Make sure the Google account has access to that file.
5. Create an OpenAI API key in your OpenAI account. In n8n, create two OpenAI credentials or reuse one for both Embeddings and Chat. Select the OpenAI credential on the Embeddings nodes and the Chat Model node.
6. Set the OpenAI Chat Model to a supported model such as gpt-4o-mini to match the template. Save the node.
7. Prepare Qdrant. If using Qdrant Cloud, copy the API URL and API Key. In n8n, create a Qdrant credential with the API URL and API Key. Save it with a clear name.
8. Open the Qdrant Vector Store node used for ingestion and select your Qdrant credential. Confirm the collection name is crowd or change it to your preferred collection.
9. Open the Qdrant Vector Store node used for retrieval and select the Qdrant credential. Decide if you want dynamic per company collections. If not needed, set a fixed collection name to match ingestion.
10. Review the text splitter settings on the loader. The chunk size is 3000 and overlap is 200. Adjust only if your documents need smaller or larger pieces.
11. Open the Webhook node. Copy the Production URL. Your request must send JSON with an input field and, if using multi tenant, a company field that matches a Qdrant collection.
12. Click Execute Workflow and trigger the Manual node once to ingest the Google Drive PDF into Qdrant. Wait for the run to complete without errors.
13. Test the query path. Send a POST request to the Webhook URL with a JSON body like { "input": "Your question", "company": "crowd" }. Confirm the Respond to Webhook node returns an answer.
14. If you see empty answers, check that the PDF was downloaded, embeddings credentials are valid, and the Qdrant collection name on retrieval matches the ingestion collection.