Artifacts-for-reproducibility-Daniel Anderson

This repository contains an automated evaluation framework designed to assess the structural adherence of microservice architectures generated by Large Language Models (LLMs). The framework combines static source-code analysis, dependency graph extraction, architecture interpretation, violation detection, and metric computation. The pipeline extracts dependency relationships from monolithic software systems using Tree-sitter and NetworkX, generating a structural graph that represents inheritance, composition, dependency injection, repository usage, and JPA relationships. These graphs are then compared against microservice decompositions proposed by LLMs under different prompting strategies, including zero-shot and few-shot approaches. The framework automatically detects structural violations caused by dependencies crossing service boundaries and computes quantitative metrics such as Dependency Violation Rate (TVD), Basic Cohesion (BC), and Granularity. The objective is to provide a reproducible method for evaluating how closely LLM-generated architectures align with the actual organization of source code. The project was developed as part of a Master's research program in Applied Artificial Intelligence and Software Engineering and is intended to support empirical studies on software architecture reconstruction, microservice migration, and AI-assisted software engineering.