What is multi-query RAG?

The Multi-query RAG template enables the integration of data from multiple sources by rephrasing queries, enhancing the generation of more accurate and contextually relevant responses.

Key features of the multi-query RAG workflow

• Generates multiple query variations to capture a broader range of data.
• Leverages multiple data sources to retrieve information.
• Enables personalized, domain-specific content generation.

Who can benefit from the multi-query RAG template?

• Businesses can improve decision-making by retrieving insights from diverse data sources.
• Customer Support Teams can provide more accurate and contextually relevant responses by integrating data from various systems.
• Internal Teams can enrich company chats by accessing knowledge stored across multiple internal sources.

How does the multi-query RAG template operate?

• Two versions of a query are generated.
• Each query is converted into vector representations (embeddings).
• The generated embeddings are used to search across multiple knowledge bases (Weaviate and Pinecone).
• A response is generated by combining information from two data sources.

Customizing the multi-query RAG workflow for your needs

• Integrate additional data sources for more comprehensive insights.
• Setup streaming for better user experience.
• Switch to alternative models for embedding or response generation. Use different storage solutions (such as PgVector, Qdrant, Chroma, Milvus).
• Update prompts to ensure the tone and style remain accurate and aligned with user needs.

Performance metrics and monitoring

• Monitor the latency and accuracy of retrieved documents, as well as the tone and correctness of the final responses, using both custom and prebuilt metrics.
• Monitor number of tokens and cost of inference.

How to use multi-query RAG template

1. Press the“Use Template” button.
2. Press the“Test” button.
3. Type your query into the “query” field.
4. Press “Run”.