International Journal of Science, Engineering and TechnologyAuthors: Grace Phillips, Alexander Brooks, Victoria Lewis, Abigail Scott, Chaitanya Srinivas, Rishi Kumar
Abstract: Large Language Models (LLMs) have emerged as transformative technologies for enhancing automation, intelligence, and decision-making across modern DevOps ecosystems. This research presents a comprehensive framework for Large Language Model–Driven Automation for DevOps Workflow Optimization through an evidence mapping and experimental analysis approach. The study systematically investigates the application of LLMs in key DevOps functions, including continuous integration and continuous deployment (CI/CD), infrastructure as code, incident management, log analysis, root cause identification, automated testing, security monitoring, and operational knowledge management. An evidence mapping methodology is employed to synthesize existing research, industrial practices, and emerging trends, providing a structured understanding of the current state of LLM adoption in DevOps environments. Furthermore, experimental evaluations are conducted to assess the effectiveness of LLM-driven automation in improving deployment efficiency, reducing manual intervention, accelerating incident resolution, enhancing workflow orchestration, and supporting intelligent decision-making. The proposed framework integrates natural language understanding, contextual reasoning, and adaptive automation capabilities to create a scalable and resilient DevOps ecosystem. Findings indicate that LLM-enabled automation significantly improves operational productivity, reduces system downtime, enhances software delivery performance, and strengthens collaboration between development and operations teams. The research contributes a practical and analytical foundation for organizations seeking to leverage generative artificial intelligence to optimize DevOps workflows while addressing challenges related to reliability, governance, security, and model transparency in enterprise-scale environments.
DOI: http://doi.org/10.5281/zenodo.20637002
