Architecture Exploration and Reflection Meet LLM-based Agents

Abstract

This work explores large-language-model-based agents for architectural design, focusing on autonomous search for patterns and tactics and reflection on the trade-offs of candidate decisions. It introduces ReArch, a proof of concept that adapts ideas from ReAct and LATS to architecture reasoning, and reports initial case-study results showing how reflective agents can support iterative exploration and assessment of design alternatives.

Publication
_2025 IEEE 22nd International Conference on Software Architecture Companion (ICSA-C), Odense, Denmark, pp. 46-50, https://doi.org/10.1109/ICSA-C65153.2025.00015_