An AI-Assisted Approach for Microservice Identification in Service-Oriented Architectures Using Large Language Models

Abstract

Identification of microservices presents multiple challenges within legacy Service-Oriented Architecture (SOA) such as tightly coupled services, implicit domain logic, and insufficient tools for automated service boundary identification. This paper introduces an AI-assisted conceptual and methodological approach to demonstrate the transition from SOA to Microservices Architecture (MSA) during the initial phases within clear design boundaries and human oversight. Architectural Reasoning Context (ARC) is the core of this approach, a machine-readable representation that combines expert-enriched metadata with expert-defined service boundary rules. The ARC guides large language models (LLMs) by providing structured prompts to generate candidate microservice groupings. This approach, executed sequentially, starts with the analysis of source code and extraction of functional units, which are annotated by domain experts. The service boundary rules are formally encoded and combined with functional unit metadata to construct the ARC representation. Using LLM-based reasoning, the proposed method processes ARC to suggest candidate microservices in JSON format, which are iteratively refined via human-in-the-loop validation, and based on the finalised JSON groupings, it produces a reviewable report for expert evaluation. After validation, to support downstream migration efforts, a final microservice identification proposal is generated. This pipeline maintains architectural integrity, emphasises explainability, policy-compliant microservice identification and expert validation, while minimising manual effort. achieving an average Net Accuracy of 66.66–70.37% and Rule Compliance of 73.80–76.66%, with consistent alignment with expert decisions, demonstrating its effectiveness.