Retrieval-Augmented Generation for Large Language Model-Based Intelligent Assistants: A Review

Abstract

Large language models have accelerated the development of intelligent assistants by providing flexible natural- language understanding and generation. However, hallucination, knowledge staleness, and limited coverage of domain-specific information continue to restrict their reliability in knowledge-intensive tasks. This review examines how Retrieval -Augmented Generation (RAG) can strengthen LLM-based intelligent assistants by connecting generative capability with external, maintainable knowledge. It synthesi zes research on the technical foundations of RAG, key components and optimization strategies, and applications and challenges in intelligent-assistant settings. The review finds that RAG can improve knowledge accuracy and timeliness by grounding responses in retrieved evidence and allowing knowledge resources to be updated independently of the base model. These benefits are conditional: unreliable retrieval, poorly maintained sources, ineffective use of context, and fragmented evaluation can still produce u nsupported or unsafe answers. Reliable deployment, therefore, requires coordinated retrieval quality, knowledge management, generation control, and trustworthy evaluation. Future RAG-based assistants should combine these capabilities to become scalable, secure, evidence-aware, and verifiable systems.