Abstract
This study proposes a hybrid semantic question-answering (QA) system for the Kazakh language that integrates ontological modeling, linguistic processing, and large language models (LLMs). The approach combines the structured reasoning of ontologies with the contextual understanding of neural language models to improve accuracy and interpretability in low-resource settings. The system includes five core modules—user interface, linguistic analyzer, LLM-based knowledge processor, ontology manager, and knowledge base—interacting through a unified architecture. A corpus of over 50,000 question–answer pairs was created for evaluation. Experimental results demonstrate significant improvements in precision, recall, F1, and mean reciprocal rank (MRR) compared with BERT-QA and ontology-only baselines. Statistical validation confirms the reliability and scalability of the hybrid model. The system is deployed as a functional web platform at The system is deployed as a functional web platform that supports semantic question answering for the Kazakh language.