📖 Adaptive RAG strategy with query classification & retrieval (Gemini & Qdrant)

⚡ 3,255 views · 📖 Internal Wiki & Knowledge Base

Description

This n8n workflow implements a version of the Adaptive Retrieval-Augmented Generation (RAG) framework. It recognizes that the best way to retrieve information often depends on the type of question asked. Instead of a one-size-fits-all approach, this workflow adapts its strategy based on the user’s query intent.

🌟 How it Works

1. Receive Query: Takes a user query as input (along with context like a chat session ID and Vector Store collection ID if used as sub-workflow).
2. Classify Query: First, the workflow classifies the query into a predefined category. This template uses four examples:
   • Factual: For specific facts.
   • Analytical: For deeper explanations or comparisons.
   • Opinion: For subjective viewpoints.
   • Contextual: For questions relying on specific background.
3. Select & Adapt Strategy: Based on the classification, it selects a corresponding strategy to prepare for information retrieval. The example strategies aim to:
   • Factual: Refine the query for precision.
   • Analytical: Break the query into sub-questions for broad coverage.
   • Opinion: Identify different viewpoints to look for.
   • Contextual: Incorporate implied or user-specific context.
4. Retrieve Info: Uses the output of the selected strategy to search the specified knowledge base (Qdrant vector store - change as needed) for relevant documents.
5. Generate Response: Constructs a response using the retrieved documents, guided by a prompt tailored to the original query type.

By adapting the retrieval strategy, this workflow aims to provide more relevant results tailored to the user’s intent.

⚙️ Usage & Flexibility

• Sub-Workflow: Designed to be called from other n8n workflows, passing user_query, chat_memory_key, and vector_store_id as inputs.
• Chat Testing: Can also be triggered directly via the n8n Chat interface for easy testing and interaction.
• Customizable Framework: The query categories (Factual, Analytical, etc.) and the associated retrieval strategies are examples. You can modify or replace them entirely to fit your specific domain or requirements.

🛠️ Requirements

• Credentials: You will need API credentials configured in your n8n instance for:
  • Google Gemini (AI Models)
  • Qdrant (Vector Store)

🔗 Nodes Used

Execute Workflow Trigger, AI Agent, Simple Memory, Summarize, Chat Trigger, Qdrant Vector Store

📥 Import

Download workflow.json and import into n8n: Workflow menu → Import from File

📖 Importing guide · 🔑 Credential setup