Large Language Models (LLMs) demonstrate increasing proficiency in solving complex tasks and explaining scientific concepts, positioning them as potential tools for democratizing access to personalized education. In low- and middle-income countries, students in rural regions often lack access to high-quality tutoring due to high costs and limited human resources, while the impact of LLMs for this setting is underexplored. To address this gap, we present a pilot study exploring the feasibility of LLMs as personalized tutors for Vietnamese K–12 students preparing for university entrance exams with 540 yes/no questions and focusing on three core STEM subjects: mathematics, physics, and chemistry. Preliminary results highlight both opportunities and challenges and offer early insights into the practicality of LLM-driven tutoring systems in resource-constrained educational environments.