Large Language Models for Hospitality Workforce Training: Design and Validation of an Intelligent Agent System

The tourism and hospitality industry relies fundamentally on the quality of human interactions, yet the sector continues to grapple with significant challenges in effectively and consistently training its workforce using resource-intensive traditional methods. This study addresses these challenges by presenting the design, development, and validation of an intelligent agent for training and evaluation, powered by Google’s Gemini 2.0 Flash model. The system processes internal organizational documentation to build a knowledge base, generates diverse question types for training, and provides automated evaluation and personalized feedback. Validation was conducted in a controlled laboratory environment corresponding to Technology Readiness Level 4 (TRL 4). The system achieved an overall success rate of approximately 82% across all test cases. It demonstrated perfect performance (100%) in social interaction and guided training capabilities. Notably, the automated evaluation engine achieved a 92% agreement rate with expert benchmarks, even for open-ended responses. However, limitations were identified in managing ambiguity and performing deep inferential reasoning beyond explicit documentation. The findings confirm the technical and functional viability of LLM-powered agents for automating hospitality training. This technology offers a scalable, objective solution that significantly reduces resource requirements while enabling personalized learning, although future optimization is needed for complex inference scenarios.