Adaptive RAG with Google Gemini & Qdrant: Context-Aware Query Answering

This n8n workflow demonstrates an advanced Retrieval Augmented Generation (RAG) system that uses Google Gemini for language modeling and Qdrant for vector storage to provide context-aware answers to user queries. It dynamically adapts its response based on the presence of relevant context.

The workflow is designed to enhance the quality and relevance of AI-generated responses by retrieving pertinent information from a vector database before generating a reply.

What it does

1. Listens for Chat Messages: The workflow is triggered by an incoming chat message, acting as the user's query.
2. Initial Query Processing: The incoming chat message is prepared for further processing.
3. Adaptive Logic: A Switch node determines the next course of action based on a predefined condition (e.g., whether a specific context is available or a certain query type is detected).
4. Contextual AI Agent (Path 1 - If context is available):
   • An AI Agent (LangChain Agent) is initialized with a Simple Memory to maintain conversational context.
   • It utilizes a Qdrant Vector Store to retrieve relevant documents based on the user's query.
   • Embeddings Google Gemini is used to convert text into vector embeddings for similarity search in Qdrant.
   • Google Gemini Chat Model is employed as the large language model (LLM) for generating responses, leveraging the retrieved context.
   • The agent processes the query, retrieves context, and generates a context-aware answer.
5. Direct AI Agent (Path 2 - If no specific context is required/available):
   • A separate AI Agent is initialized with a Simple Memory.
   • It directly uses the Google Gemini Chat Model to generate a response without explicit vector store retrieval, suitable for general queries or when no specific context is found.
6. Summarize Response: The final response from either AI agent path is summarized.
7. Responds to Webhook: The summarized AI-generated answer is sent back as a response to the original chat message trigger.

Prerequisites/Requirements

• n8n Instance: A running n8n instance.
• Google Gemini API Key: For the Embeddings Google Gemini and Google Gemini Chat Model nodes.
• Qdrant Instance: Access to a Qdrant vector database instance (either self-hosted or cloud-based).
• Qdrant Collection: A pre-populated Qdrant collection with relevant data, indexed using Google Gemini embeddings, for the Qdrant Vector Store node to function effectively.
• Chat Platform: A chat platform configured to send messages to n8n via a webhook (e.g., Slack, Telegram, custom frontend).

Setup/Usage

1. Import the Workflow:
   • Download the provided JSON file.
   • In your n8n instance, go to "Workflows" and click "New".
   • Click the three dots next to "Save" and select "Import from JSON".
   • Paste the workflow JSON content or upload the file.
2. Configure Credentials:
   • Locate the Embeddings Google Gemini and Google Gemini Chat Model nodes.
   • Set up a new credential for Google Gemini by providing your API Key.
   • Locate the Qdrant Vector Store node.
   • Set up a new credential for Qdrant by providing your Qdrant API Key and Host.
3. Configure Qdrant Collection:
   • In the Qdrant Vector Store node, specify the name of your Qdrant collection. Ensure this collection exists and contains the data you want to use for RAG.
4. Configure Chat Trigger:
   • The When chat message received node acts as the entry point. After activating the workflow, n8n will provide a webhook URL. Configure your chat platform to send messages to this URL.
5. Customize Adaptive Logic (Optional):
   • Adjust the condition in the Switch node (e.g., {{ $json.query.includes('context') }}) to define when the workflow should use the Qdrant-backed RAG path versus the direct LLM path.
6. Activate the Workflow: Once all credentials and configurations are set, activate the workflow.

Now, when a chat message is received, the workflow will intelligently decide whether to retrieve context from Qdrant before generating a response with Google Gemini, providing a more adaptive and informed answering system.