Model-based software engineering (MBSE) utilizes high-level abstractions to streamline discussions, facilitate code generation, and validate system designs, with models acting as essential blueprints for developers. However, as the complexity of these models increases, identifying optimal options for stakeholders becomes increasingly challenging. This paper introduces ADVICE, an innovative human-in-the-loop AI tool designed to learn stakeholder preferences from textual inputs. Leveraging large language models, ADVICE explores and infers human preferences, integrating multi-objective swarm optimization to identify solutions that align with domain-specific constraints. We outline the iterative development of ADVICE and its functionality in expediting stakeholder identification, exploration, assessment, and consensus on critical model options. The effectiveness of ADVICE will be evaluated based on its ability to enhance system performance by implementing agreed-upon options. A practical example is provided, illustrating how drone fleet operators can utilize ADVICE to swiftly detect and mitigate dangerous scenarios. Our findings demonstrate the potential of ADVICE to significantly improve decision-making processes in model-based software engineering, ultimately leading to more robust and efficient system designs.