Towards an agent-based framework for guiding design exploration.

One of the premises of conceptual design is that the designer must evaluate a range of candidate solutions before selecting the final solution. Tool support is critical to aid designers in that exploration, because the design space is usually large and involves multiple constraints. A modality of assistance is that the tool criticizes the current design and provides the designer with recommendations for improving it. Traditional knowledge-based systems and optimization tools tend to be inappropriate when the designer is actively involved in the search loop, because the design proposals should match the designer's context. To address this challenge, this paper describes an agent-based framework for developing design recommendation tools that help designers to perform explorative search more effectively. The approach is exemplified with an experimental design assistant for software architecture design.