Under the AI4S(AI for Science)paradigm,knowledge services in academic libraries are undergoing a profound intelligent transformation,making the development of disciplinary vertical knowledge service systems an increasingly urgent priority.Focusing on marine-related disciplines,this paper proposes a construction logic for such systems and develops a framework characterized by resource-driven development,scenario-based architecture,collaborative construction,and tiered implementation.Taking fisheries science—an A+discipline at Ocean University of China—as an exemplary case,it established a tripartite collaboration among the library,subject experts,and the technical team to iteratively co-construct an AI knowledge base for fisheries science.The system adopts a four-layer architecture comprising content,knowledge,model,and application layers,underpinned by an overarching security assurance system and supported by multi-level knowledge services tailored to three-level disciplines.The content layer integrates multi-source heterogeneous data into three types of marine-specific corpora;the knowledge layer constructs a computable and traceable disciplinary knowledge graph;the model layer couples intelligent agents with retrieval-augmented generation to mitigate hallucination;and the application layer delivers differentiated services across teaching,research,and industry scenarios.The system has been empirically validated in teaching,research,and industrial service contexts.Comparative testing against several mainstream general-purpose large language models and CNKI AI demonstrates that the fisheries AI knowledge base returns a substantially higher proportion of academic and core authoritative resource,including a notably greater share of foreign-language sources,thereby outperforming general-purpose tools in authority,relevance,and traceability.In teaching,the system supports lesson preparation,classroom instruction,and after-class tutoring,yielding high learner satisfaction in pilot courses.In research,it embeds resource investigation into the entire research workflow,markedly improving efficiency.In discipline-level decision-making,it enables frontier monitoring,competitor benchmarking,and patent risk warning with traceable evidence chains.The findings suggest that the core competitiveness of discipline-specific knowledge services derives from the deep coupling of authoritative disciplinary data and trustworthy intelligent technologies,while their long-term value appreciation depends on the discipline's professional foundation,demand orientation,and the continuous interaction among smart librarian teams and multi-agent systems.By distilling transferable principles from a single case,this paper offers a reference for the intelligent transformation of knowledge services in academic libraries in the AI4S era.