<![CDATA[The development of generative artificial intelligence has opened opportunities for strengthening Islamic legal learning through a digital forum that remains connected to authoritative textual references. This study aims to design and build a web-based bahtsul masail discussion system using a multi-agent AI architecture and the Retrieval-Augmented Generation (RAG) method. The proposed system simulates the deliberative roles commonly found in pesantren-based Islamic legal discussion, namely Moderator, Mubahits, Mu'aridh, and Mushahih. The research applies a Research and Development approach with the Waterfall model, covering requirement analysis, interface design, multi-agent workflow design, implementation, functional testing, and evaluation. RAG is implemented by allowing users to upload PDF documents of kitab kuning and assign the documents to particular agent roles. The uploaded texts are then used as contextual grounding so that each agent can formulate arguments, rebuttals, and final decisions based on traceable references rather than unsupported model memory. The application is implemented using HTML, CSS, and JavaScript on the front end, while the AI reasoning process is orchestrated through an API-based large language model. Functional testing shows that the system can complete five sequential stages of bahtsul masail, display role-based responses, manage uploaded references, and present discussion history. The main contribution of this study is a system design that combines pesantren deliberation procedures with AI-based retrieval support to provide an interactive learning medium for Islamic law, while emphasizing that the final authority of legal validation remains with qualified scholars]]>
