Helping Novice Architects to Make Quality Design Decisions Using an LLM-Based Assistant

Abstract

Architectural knowledge and specifically design decisions have become first-class entities to be captured routinely in a design process. However, the quality of the decisions captured is often low. Part of the problem is that reflections intended to criticize, and thus improve, the decisions are seldom made in architecture teams, particularly when involving novice architects. To improve reflective practices and capture better decisions, we propose an design assistant approach based on generative AI techniques. Our assistant, called ArchMind, relies on two information sources: architectural knowledge about patterns, and information about the system under design. Furthermore, the assistant takes advantage of LLMs to progressively aid users in selecting and assessing alternative decisions, until capturing them using an Architecture Decision Record format. ArchMind mainly targets novice architects as discussed in an initial experiment using the assistant off-line for a classroom project. The generated ADRs were concrete and well-justified in their design rationale, although they tended to miss system-specific details.