This paper presents the development of an enhanced, resource-efficient artificial intelligence (AI) tool for generating domain-specific, etiquette-compliant responses to automate airport customer service operations. While pretrained large language models (LLMs) can generate fluent text, their direct application often produces generic, unstructured, and domain-misaligned outputs lacking the politeness and professionalism required in service contexts. To address these limitations, we curated a labeled dataset from online resources and synthetic generations, ensuring balanced category representation and clear separation between high- and low-quality responses, and used it for supervised fine-tuning of the LLM to enhance the generation. Our approach employs the open-source GPT-2 to ensure transparency, cost-effectiveness, and adaptability, fine-tuned through Low-Rank Adaptation (LoRA) for parameter-efficient, resource-friendly training. The fine-tuned model was evaluated against its pretrained counterpart using three complementary methods: (1) lexical and semantic similarity with reference answers via ROUGE and BLEU scores, (2) label-consistency verification using four independent classifiers—ModernBERT-base, e5-small-v2, DeBERTa-v3-large, and RoBERTa-base, and (3) visual evaluation through an interactive demonstration interface for qualitative comparison. Results demonstrate that the fine-tuned model significantly surpasses the pretrained model in contextual alignment, clarity, politeness, and label accuracy. Beyond airport operations, this framework can be applied to any customer service or agentic environment where response generation must balance technical accuracy, domain expertise, and alignment with human values.