Adaptive RAG Query Classification Workflow

Target Audience

• Data Scientists: Those looking to enhance their data retrieval processes with adaptive strategies.
• Developers: Individuals who want to integrate advanced query classification and retrieval mechanisms into their applications.
• Business Analysts: Professionals needing precise information retrieval for analysis and reporting.
• Researchers: Academics and scholars seeking comprehensive insights from varied sources.
• Customer Support Teams: Teams requiring contextual understanding to provide accurate responses to user queries.

Problem Solved

This workflow addresses the challenge of efficiently retrieving relevant information based on user queries by classifying them into four distinct categories: Factual, Analytical, Opinion, and Contextual. It enhances the retrieval process by applying tailored strategies for each query type, ensuring that users receive the most accurate and relevant responses. This adaptive approach significantly improves the quality of information retrieval, leading to better user satisfaction and decision-making.

Workflow Steps

1. Input Trigger: The workflow initiates through a webhook or chat interface, expecting inputs such as user_query, chat_memory_key, and vector_store_id.
2. Combine Inputs: The Combined Fields node standardizes the inputs for further processing.
3. Query Classification: The Query Classification node classifies the user_query into one of four categories using a Google Gemini agent.
4. Adaptive Strategy Routing: A Switch node directs the flow based on the classification result.
5. Strategy Implementation: Depending on the classification:
   • Factual: Enhances the query for precision.
   • Analytical: Breaks the query into sub-questions for comprehensive coverage.
   • Opinion: Identifies diverse perspectives on the topic.
   • Contextual: Infers relevant implied context.
6. Set Prompt and Output: Prepares the output from the strategy step and a tailored prompt for the final answer generation.
7. Document Retrieval: Retrieves the most relevant documents from the Qdrant vector store using the adapted query.
8. Context Preparation: Concatenates the retrieved document contents to form a cohesive context.
9. Answer Generation: The final response is generated using the tailored prompt and concatenated context.
10. Response: Sends the generated answer back to the user via the webhook.