[Objectives] This study aims to explore the feasibility of employing DeepSeek, a large language model, to promote intelligent hydrological analysis through its natural language interaction and code generation functions. This research innovatively applies DeepSeek to engineering hydrological analysis, promoting intelligent development in the field of engineering hydrology. [Methods] First, based on the core concepts and characteristics of engineering hydrology discipline, it was concluded that DeepSeek’s application scenarios such as code generation, code rewriting, and code explanation were highly suitable for engineering hydrology, a field heavily dependent on data. Focusing on the typical task of frequency analysis of hydrological data, this study used a case-driven method and designed a two-stage experiment. During the data cleaning phase, daily water level data incorporating compound water level recording methods were fed into the system, and MATLAB cleaning code was iteratively generated using structured prompts. In the data analysis phase, the annual maximum water levels, 3-day and 5-day moving average maximum sequences during the flood season were generated, and the Pearson Type III (P-III) distribution was used to calculate key frequency design values such as 1% and 5%. Finally, a quantitative comparison was conducted between DeepSeek’s calculated results and conventional eye-fitting curve outcomes to evaluate the accuracy of the results. [Results] In terms of efficiency, the processing time for multiple prompts ranged from 33 to 109 seconds. Standardized tasks (such as moving average calculations) achieved “prompt as code”, substantially reducing programming time and significantly enhancing workflow efficiency. Additionally, the automated optimization of existing inefficient code notably improved efficiency. Regarding accuracy, DeepSeek could accurately identify user requirements and precisely interpret professional concepts. It achieved a 100% accuracy rate in the first attempt when interpreting key concepts such as the P-III distribution and flood season averages. However, for low-frequency terms (e.g., compound recording method), 2-3 rounds of prompt iteration were required. Additionally, DeepSeek’s calculated average and C_v parameters were consistent with those obtained using conventional methods, further demonstrating its high precision. [Conclusions] DeepSeek significantly lowers the technical barriers to engineering hydrological analysis. Its natural language interaction capability serves as an “intelligent bridge” between professional requirements and code implementation, while its automated data processing and model calculation alleviate practitioners’ workload, promoting the integration of AI technology from academic research into engineering practice. In the future, with in-depth research and expanded applications, DeepSeek is expected to evolve from an auxiliary tool into a core engine driving the transformation of engineering hydrology from “experience-based decision-making” to “knowledge-data collaborative decision-making,” thereby providing foundational support for intelligent water conservancy.