Systematic Technical Survey on LLMOps: Lifecycle, Tools, Challenges, and Emerging Practices
Year: 2025
Type: dissertation
Study design: Systematic Review
Abstract: The emergence of Large Language Models (LLMs) has transformed artificial intelligence applications across industries, yet their operational management presents challenges that exceed traditional Machine Learning Operations (MLOps) capabilities, creating the need for specialized frameworks.This thesis presents a systematic technical survey of LLMOps through comprehensive literature analysis of 117 academic and industry sources from 2020-2025, addressing three research questions: identifying distinctive components differentiating LLMOps lifecycle from traditional MLOps, categorizing unique operational challenges in LLM deployment, and assessing current mitigation strategy effectiveness.One of the main contributions is developing an integrated literature-informed LLMOps operational framework that synthesizes scattered knowledge into a comprehensive six-phase lifecycle model encompassing prompt engineering, data preparation, model adaptation, evaluation, deployment, and continuous improvement with human feedback.Using thematic synthesis and systems thinking analysis, the research identifies four primary operational challenge categories: evaluation complexity, economic constraints, data management difficulties, and integration complications from fragmented tooling ecosystems.The investigation reveals that these challenges exhibit systemic interdependencies requiring holistic management approaches rather than isolated optimization strategies, leading to the development of a systems interdependence framework explaining LLMOps effectiveness through dynamic tension management between human expertise, economic constraints, and adaptive capabilities.Key contributions include establishing LLMOps as a distinct operational discipline, providing structured theoretical foundations, and demonstrating that successful implementation requires simultaneous development of evaluation methodologies, economic optimization approaches, and integration protocols, thereby offering practical guidance for organizations and research priorities for advancing this emerging field. I want to extend my deepest thanks to my dear wife and life
Author: Özer, Furkan
Institution: University of Eastern Finland
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7215572191