The rapid rise of artificial intelligence (AI) has profoundly transformed software architectures, placing machine learning components at the core of modern systems. However, monolithic architectures, long considered the standard, quickly reach their limits when faced with the demands of scalability, maintainability, and continuous evolution inherent to AI pipelines, which include critical stages such as data preprocessing, training, and model deployment. Traditional migration approaches toward microservices, originally designed for systems without intelligent components, prove inadequate in this context, as they fail to capture the complexity and interdependencies of data flows and learning processes.
To address these limitations, we propose a decomposition approach explicitly tailored to monolithic AI-based systems. It brings together two complementary aspects : first, a review that highlights the strengths and limitations of existing approaches for transforming monolithic systems into microservices; and second, an original decomposition approach that combines the use of architectural patterns with the assistance of large language models to identify and group AI pipeline components into cohesive and loosely coupled microservices.
We validate our approach on three monolithic AI-based systems and compare our decomposition results with two baseline approaches from the literature. The results demonstrate the effectiveness of our method in producing modular and AI-aware decompositions, with a precision of 84% and a recall of 65%, outperforming the baseline approaches.
| Date | 9 Mar 2026 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Manel Abdellatif (Supervisor) & Naouel Moha (Co-supervisor) |
|---|
Ghlissi, H. (Author),
Abdellatif (Supervisor) &
Moha (Co-supervisor),
9 Mar 2026Student thesis: Master's thesis › Master in Engineering: Engineering