Graph Neural-Network SLA Regression Detection Across MLOps and DataOps Workflows

Service Level Agreements (SLAs) are critical in ensuring reliability, performance, and compliance within modern AI-driven operations such as MLOps (Machine Learning Operations) and DataOps (Data Operations). However, SLA regressions—unexpected degradations in system performance or behavior—are challenging to detect due to the complexity and dynamic nature of workflow dependencies. This paper proposes a novel approach that leverages Graph Neural Networks (GNNs) to model the interconnected components of MLOps and DataOps pipelines as dynamic graphs. By learning node and edge embeddings from historical operational data, our method detects anomalous patterns indicative of SLA regressions in real time. We validate our model on multiple real-world datasets and demonstrate superior performance compared to traditional rule-based and statistical baselines. Our findings suggest that GNN-based regression detection offers scalable, interpretable, and proactive SLA monitoring across diverse AI pipelines.