Design and Prototype Implementation of a Modular AI ModelOps Platform for End-to-End Machine Learning Lifecycle Management

Machine learning models are often developed through experiments and notebooks, but moving them into reliable production systems requires coordinated data management, experiment tracking, model versioning, deployment, and continuous monitoring. Disconnected tools and manual release steps can reduce reproducibility and make model updates difficult to govern. This paper presents the design and prototype implementation of a modular AI ModelOps Platform that brings key lifecycle operations into a common system. The platform combines a React and TypeScript dashboard, FastAPI-based control-plane APIs, PostgreSQL application metadata, Redis-backed background services, MLflow experiment tracking and model registry, a model router, and Docker-based model serving. Its delivery workflow uses GitHub Actions for validation and image building, GitHub Container Registry (GHCR) for versioned artifacts, and a pull-based staging release checker on an Ubuntu host. Before promotion, a candidate image is checked for source and version metadata and tested through health and prediction requests; rollback safeguards preserve the previous deployment when validation fails. Functional verification has confirmed a staging prediction from the registered demo-classifier model version 5. Drift analysis and governed retraining are included in the intended lifecycle feedback design, while fully autonomous retraining is treated as an extension requiring further validation. The paper presents the system architecture, implementation workflow, related research, evaluation limits, and future development directions