Operationalizing AI at Scale: A Cloud-Agnostic Framework for Infrastructure, Deployment, and MLOps in Enterprise AI Solutions
Year: 2026
Type: dissertation
Abstract: Organizations can now develop artificial intelligence prototypes quickly. The harder task begins when those prototypes have to run in production: integrating them with existing systems, keeping costs and risks under control, and managing change over time. This transition from an initial experiment to an operational service still often depends on manual configuration, provider-specific cloud services, and uneven processes for deployment, monitoring, and model management. This dissertation treats that transition as a software engineering problem and proposes a cloud-agnostic framework for operationalizing AI in enterprise environments. Its purpose is not to replace existing MLOps platforms, but to organize engineering decisions that recur across projects: infrastructure preparation, artifact promotion between environments, service and model observation, and traceability across the solution lifecycle. The framework separates operational principles from provider-specific implementation choices, seeking to reduce cloud lock-in without giving up the automation offered by modern infrastructure. The work combines a systematic literature review, requirements specification, architectural decisions, and a reference implementation based on Kubernetes, GitOps, Kubeflow, MLflow, Argo Workflows, Argo Events, Crossplane, Vault, and External Secrets. Validation links use cases, requirements, architecture phases, and repository artifacts, and includes the reference ml-fraud-detection project. The evaluation considers portability, reproducibility, operational reliability, observability, and alignment with enterprise requirements. Long-running production validation and complete provider parity fall outside the scope of this work. The result is a framework, supported by a reference implementation, that helps teams move AI solutions into production while preserving room to adapt tools and services to their technical context.
Author: Nuno Oliveira Fernandes
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7202065031