Multiple Abstraction Level RAG (MALRAG) improves on traditional Retrieval-Augmented Generation by retrieving information at different levels of abstraction, providing more comprehensive and contextually appropriate responses.

• Creates a hierarchical knowledge structure from textbooks to enable retrieval at multiple abstraction levels
• Demonstrates superior performance compared to conventional RAG and fine-tuning approaches
• Achieves more nuanced responses by leveraging both high-level concepts and specific details
• Proves particularly effective in specialized domains like Glycoscience

This research has significant implications for medical applications by enabling more precise and medically-accurate responses in specialized healthcare domains, while maintaining the flexibility to adapt to new medical knowledge without expensive retraining.

Multiple Abstraction Level Retrieve Augment Generation