Creative synthesis of indoor scenes that satisfy personalized aesthetic preferences presents significant challenges due to the subjective nature of creativity and the complexity of the design space. While existing scene synthesis methods primarily address functional and ergonomic considerations, they often overlook creative, individualized design aspects. To bridge this gap, we propose a novel framework that leverages Large Language Models (LLMs) to creatively synthesize indoor scenes through interactive evolutionary algorithms. Our method dynamically captures and refines user-specific preferences via iterative human feedback, guided by the semantic comprehension and generative capabilities of LLMs. Additionally, we introduce an adaptive camera path-planning strategy to effectively visualize generated scenes, enabling comprehensive exploration and evaluation. Extensive experiments and user studies demonstrate that our LLM-driven creative synthesis method significantly surpasses traditional evolutionary approaches in scene plausibility. Our results highlight the potential of integrating human-centered LLM-based techniques into computational creativity, opening new possibilities for personalized interior scene generation and interactive visual exploration.