This study presents an intelligent internet promotion recommendation system designed to address the inaccuracy of chatbots that rely solely on large language models. The system integrates Retrieval-Augmented Generation (RAG) with an AI Agent to retrieve domain-specific knowledge and generate accurate responses. It is implemented using FastAPI and PostgreSQL to support web-based conversational interactions.

       The system was evaluated using 50 conversation cases with both AI Judge and rule-based evaluation. The results indicate good to very good performance, achieving scores of 4.60 for politeness, 4.28 for relevance, and 4.98 for readability. These findings demonstrate that the system can provide accurate, clear, and user-friendly recommendations suitable for customer service applications.