Article Details

Abstract

In response to issues in grassroots Party building work in the new era, such as fragmented knowledge management, low query efficiency, and stringent data security requirements, this study designs a secure, intelligent, and efficient knowledge management and Q&A system for Party building. Adopting a local-first hybrid intelligent agent approach, it leverages low-code AI development platforms such as Dify and Coze, and integrates key technologies including DeepSeek large language model, BGE-M3 embedding model, MinIO object storage, and PaddleOCR to construct a system. This system can parse documents in multiple formats, perform intelligent categorization, conduct semantic search, and achieve precise information traceability, forming an intelligent Party building knowledge base system. The system follows a standardized three-layer architecture—comprising the data access layer, core processing layer, and user interface layer—and supports access via both PC and H5 mobile devices. The proposed framework effectively integrates cutting-edge artificial intelligence technologies with the requirements of Party and government affairs, ensuring data security through localized deployment.

Keywords

agent; DeepSeek; Dify; Coze; Retrieval-Augmented Generation; knowledge base; BGE-M3

References

1. [1] Chen Yang, Li Yunfeng, Mou Fuquan, etc The Application of Large Language Models in Earth Sciences: A Case Study of Retrieval Enhancement Generation System Based on Deepseek-R1 [J]. Geological Review, 2025, 71 (S1): 343-344
2. [2] Zhang Jie, Wang Chao, Tang Xiong, etc A method for constructing perception decision action chain based on multi-agent: CN202510452113.3 [P]. 20225-05-23
3. [3] Yang Jun, Li Changjian Localization deployment application of DeepSeek big model: functional inspection, emerging risks, and governance paths [J]. United Front Studies Research, 2025, 9 (3): 69-83
4. [4] Wang Bochao, Feng Yinhui, Wang Yongtao Intelligent Q&A System for Fault Diagnosis of Comprehensive Mining Face Equipment Based on LLM and RAG [J]. Coal Technology, 2025, 44 (9): 207-211
5. [5] An Lianchao, Huang Yiwei, Xu Chaodong, etc Cognitive Behavioral Psychological Intervention System Based on Dify Workflow Engine: CN202510685068.6 [P]. 20225-05-26
6. [6] Chen Lulu Research on Multi label Text Classification Method for Legal Documents [D]. Shenyang: Shenyang University of Technology, 2025
7. [7] Yuan Lijun, Li Youhua, Yu Weidong Research on the Architecture of Intelligent Fusion Distributed Storage System Based on MinIO [J]. Science and Technology Bulletin, 2026, 42 (4): 38-45
8. [8] Liu Qi, Zhang Shuo, Wang Ying Research on AI exclusive Q&A agent for microcontroller courses based on the COZE platform [J]. Journal of Jilin University of Chemical Technology, 2025, 42 (12): 22-27
9. [9] Lu Shasha, Xiao Haili, Wang Xiaoning Application of Container Technology in High Performance Computing Environment [J]. Frontiers of Data and Computing Development, 2021, 3 (6): 118-126
10. [10] Chen Jie, Ye Yongfei, Kuang Shilei, etc A method, device, and computer-readable storage medium for account login: CN20241152375.1 [P]. 2022-10-29

About the authors

Tao Yangming (1987-), male, Han ethnicity, from Luohe, Henan, intermediate professional title, lecturer, assistant to the dean of the School of Artificial Intelligence, master's student, research interests include AI algorithms and natural language processing.