Empowering Smart Retail: Leveraging Large Language Models for Intelligent Shopping Assistants

Abstract

The rapid evolution of smart retail has created a growing demand for intelligent, responsive, and personalized shopping experiences. This study explores the integration of Large Language Models (LLMs), particularly GPT-based architectures, into retail ecosystems to power intelligent shopping assistants. A domain-specific LLM was fine-tuned and deployed in a simulated retail environment, where it handled natural language queries, offered product recommendations, and maintained multi-turn conversations. The system was evaluated across five key dimensions: response accuracy, personalization, context retention, response time, and user satisfaction. Results showed an average response accuracy of 91.3%, with strong personalization alignment (87.1%) and over 96% context retention in multi-turn dialogues. User surveys indicated high satisfaction with interaction quality, ease of use, and recommendation relevance. Compared to traditional rule-based systems, the LLM assistant demonstrated superior performance in contextual understanding and user engagement, albeit with a marginal increase in response time. These findings highlight the viability of LLMs as a foundation for scalable, intelligent customer service in retail. The study concludes by emphasizing the importance of ethical deployment and future optimization to enhance accessibility and real-time performance.