This work presents the design and pilot evaluation of a Retrieval-Augmented Generation (RAG) virtual assistant powered by the quantized LLaMA-7B-Spanish.Q4 model, aimed at supporting learning in international trade. Built from curated question–answer pairs on Ecuadorian customs regulations, the system integrates SBERT embeddings and FAISS retrieval to provide context-aware responses. A pilot test with 16 undergraduate students assessed clarity, usefulness, and accuracy, with results indicating positive acceptance and potential pedagogical value. While the small, homogeneous sample limits generalizability, preliminary insights and informal comparisons with keyword search suggest advantages worth exploring in larger-scale evaluations.