Design of a smart blood donation assistant based on large language model
10.13303/j.cjbt.issn.1004-549x.2026.02.012
- VernacularTitle:基于大语言模型的智慧献血服务小助手设计
- Author:
Lan LUO
1
;
Kanglie WAN
1
;
Yue ZHENG
1
;
Xiaoya ZHAO
1
;
Zhedong HAN
1
Author Information
1. Zhejiang Blood Center, Hangzhou 310052, China
- Publication Type:Journal Article
- Keywords:
blood donor services;
transfusion informatization;
large language model;
retrieval-augmented generation;
model context protocol
- From:
Chinese Journal of Blood Transfusion
2026;39(2):241-247
- CountryChina
- Language:Chinese
-
Abstract:
Objective: To develop a smart blood donation service assistant for popularizing donation-related knowledge to blood donors via intelligent Q&A support, thereby enabling precise service delivery. Methods: Based on the operational scenarios of the Zhejiang Provincial Blood Center, the system utilized the open-source Dify platform for agent orchestration, and integrated with the DeepSeek model as the language processing engine to support online real-time interaction. External tools, including the Amap API and MySQL database queries, were encapsulated via the Model Context Protocol (MCP). A professional blood knowledge base for Retrieval-Augmented Generation (RAG) was constructed using the BGE-M3 embedding model. An innovative dual-large language model collaborative verification mechanism was introduced to design the overall framework. The system was deployed privately using Docker containerization, and offline closed-loop optimization was achieved through customized Python scripts. Results: An interactive interface for blood donors was developed by integrating the chatflow Web component from Dify. The intelligent assistant Agent can recommend optimal blood donation sites and navigation routes by invoking the Amap API based on the donor's location. The Blood Donation Knowledge Agent enables timely responses to inquiries, along with reasonable suggestions and reminders. This agent specializes in the field of voluntary blood donation, empowering the assistant to answer doubts and questions for blood donors in the form of intelligent question-and-answer interaction. It also guides users through preliminary self-assessments, helping potential donors identify eligibility issues beforehand, thereby effectively increasing the on-site success rate of blood donation and reducing resource waste. Conclusion: The smart blood donation assistant validates the feasibility of the "Dify+MCP+RAG" technical architecture within the blood transfusion informatization field. The assistant not only improves the service experience for blood donors, but also, ensures the sustainable evolution of the system through its modular design and closed-loop optimization mechanism, thus providing valuable insights for the intelligent transformation of traditional blood donation service systems.
