MedeepRAG: A Retrieval-Augmented Generation System for Medical Q&A Using DeepSeek-R1
He Huang · 2025
Large Language Models (LLMs) represent a major breakthrough in Artificial Intelligence (AI), with models like DeepSeek from China demonstrating effectiveness in various natural language processing (NLP) tasks. These models, trained on large datasets, capture the intricate relationships between words in textual data. Retrieval-Augmented Generation (RAG) is a novel approach that combines retrieval-based and generation-based models to enhance text generation quality. However, LLMs have yet to achieve optimal performance in biomedical tasks, where domain-specific expertise is crucial. To address this gap, this paper proposes MedeepRAG, a model built on the distilled version of DeepSeek-R1-Distill-1.5B. MedeepRAG is designed to balance inference power and deployment efficiency with only 1.5 billion parameters. It is trained on the Medical-R1-Distill-Data-Chinese, a high-quality, structurally labeled Chinese medical dataset. Experimental results show that MedeepRAG produces responses with more reliable medical knowledge, supporting the integration of Artificial Intelligent (AI) into the medical field for enhanced Question and Answer (Q&A) in clinical settings. Future work will focus on clinical validation, multimodal data processing, and improving nested medical concept handling.