Conversational AI for primary education: Design of a culturally-aware LLM assistant

This paper presents AI Guguță, a culturally aware conversational assistant designed for primary education in Moldova. The system combines hybrid large language model pipelines with Retrieval-Augmented Generation (RAG) to provide curriculum-grounded and age-appropriate tutoring for Romanian language and mathematics. Unlike conventional educational chatbots, AI Guguță integrates cultural adaptation through a pedagogical persona inspired by Moldovan children’s literature together with collections of traditional Moldovan proverbs. The assistant further employs Socratic-style conversational strategies that encourage guided reasoning and critical thinking rather than direct answer generation. The paper describes the system architecture, RAG workflow, pedagogical design principles, and conversational interaction mechanisms underlying the assistant. The proposed framework demonstrates the feasibility of culturally grounded AI tutoring systems for primary education in low-resource linguistic and educational contexts.