Architectural Approaches and Trends in GenAI Systems for Automated Interactive Dashboard Generation: A Systematic Review

This article presents a systematic analysis of the architectural and algorithmic foundations of Generative Artificial Intelligence (GenAI) systems designed for the autonomous synthesis of interactive dashboards. The review focuses on the transition from manual, code-intensive Business Intelligence (BI) workflows toward conversational, agentic paradigms enabled by Large Language Models (LLMs). It explores the 'Analysis-Presentation Decoupling' principle and the use of structured Intermediate Representations (IR) to enhance structural controllability. Multi-agent frameworks, involving Planner, Coder, and Critic agents for iterative self-reflection, have been analyzed. Based on a comparison of modern systems such as SiriusBI, ChatVis, and NL2Dashboard, the critical role of semantic layers and the Model Context Protocol (MCP) in ensuring analytical faithfulness has been identified. The review concludes by highlighting the remaining necessity of human-in-the-loop protocols for design nuance.