n8n Document RAG System with Kimi K2, Gemini Embeddings, and Qdrant

This n8n workflow demonstrates a robust Retrieval Augmented Generation (RAG) system for document processing. It leverages Kimi K2 for document extraction, Google Gemini for embeddings, and Qdrant for vector storage and retrieval. This system allows you to ingest documents, generate vector embeddings for their content, store them in a vector database, and then use an AI Agent to answer questions based on the retrieved document segments.

What it does

This workflow automates the following steps:

1. Triggers on Manual Execution or Sub-workflow Call: The workflow can be initiated manually or by another n8n workflow calling it as a sub-workflow.
2. Receives Document Data: It expects to receive binary file data representing a document.
3. Extracts Text from Document: It uses the "Extract from File" node to parse the content of the incoming document (e.g., PDF, DOCX, TXT) into raw text.
4. Splits Text into Chunks: The extracted text is then split into smaller, manageable chunks suitable for embedding and retrieval.
5. Generates Embeddings (Sub-workflow): For each text chunk, it calls a sub-workflow (presumably named "Embeddings Sub-workflow" or similar, though not explicitly defined in this JSON) to generate vector embeddings using Google Gemini.
6. Stores Embeddings in Qdrant (HTTP Request): The generated embeddings, along with their corresponding text chunks, are then sent via an HTTP Request to a Qdrant vector database for storage and indexing.
7. AI Agent for Querying: The workflow includes an AI Agent that can receive chat messages. This agent is configured to use a "Call n8n Workflow Tool" (likely the embedding/retrieval sub-workflow) to query the Qdrant database and retrieve relevant document segments.
8. Generates Responses with Google Gemini: The AI Agent then uses the Google Gemini Chat Model to formulate a response based on the retrieved information and the user's query.

Prerequisites/Requirements

To use this workflow, you will need:

• n8n Instance: A running n8n instance.
• Kimi K2 API Access: For document extraction (implied by the directory name, though not explicit in the JSON, the "Extract from File" node handles various formats).
• Google Gemini API Key: For generating text embeddings and chat model interactions.
• Qdrant Instance: A running Qdrant vector database instance to store and retrieve document embeddings.
• Credentials Configuration: Appropriate n8n credentials configured for Google Gemini and Qdrant.
• Sub-workflow for Embeddings: A separate n8n sub-workflow configured to handle the embedding generation and Qdrant interaction (referenced by the "Execute Sub-workflow" and "Call n8n Workflow Tool" nodes).

Setup/Usage

1. Import the Workflow: Import the provided JSON into your n8n instance.
2. Configure Credentials:
   • Set up your Google Gemini API key as an n8n credential.
   • Ensure your Qdrant connection details (API endpoint, API key if applicable) are correctly configured in the HTTP Request node or as an n8n credential used by the sub-workflow.
3. Configure Sub-workflow: Create and configure the sub-workflow responsible for generating embeddings and interacting with Qdrant. This sub-workflow should be callable by the "Execute Sub-workflow" node and the "Call n8n Workflow Tool".
4. Activate Workflow: Enable the workflow in n8n.
5. Trigger the Workflow:
   • Manual: Click "Execute Workflow" in the n8n editor to test.
   • Programmatic: Send binary file data to the "Execute Workflow Trigger" node from another workflow.
   • Chat: Interact with the AI Agent via the "Chat Trigger" to ask questions about your documents after they have been processed and indexed.
6. Monitor Execution: Observe the workflow execution in n8n to ensure documents are processed, embeddings are generated, and data is stored in Qdrant correctly.