DevOps for Generative AI Systems: Continuous Delivery and Governance for Model-Integrated Applications

Abstract

Generative AI systems introduce new operational risks including model drift, prompt injection, data leakage, and unpredictable output behavior. This research proposes a DevOps operating model for model-integrated applications that unifies CI/CD, evaluation gating, safety policies, and observability for both application and model components. The framework defines deployment pipelines with automated prompt tests, model performance thresholds, and audit-ready governance for changes in prompts, retrieval sources, and models. Results show improved release safety and traceability for GenAI-enabled services while maintaining rapid iteration cycles.