Retrieval-Augmented Generation (RAG) for Reliable AI Question Answering

This paper presents a review of Retrieval-Augmented Generation (RAG) for reliable AI question answering. Large Language Models (LLMs) can sometimes generate incorrect, unsupported, or outdated information. RAG addresses these limitations by retrieving relevant information from external knowledge sources and using it to generate context-aware answers. This paper explains the fundamental architecture and working principles of RAG, including document preprocessing, embedding generation, vector-based retrieval, context construction, and answer generation. It also discusses the proposed RAG framework and evaluation metrics such as answer accuracy, faithfulness, context precision, context recall, and response latency. Furthermore, the paper examines practical applications of RAG in education, healthcare information, customer support, enterprise knowledge management, and research assistance. It discusses key challenges, including retrieval errors, hallucinations, data quality, security, and computational cost. Finally, it highlights future research directions for improving retrieval quality, evidence grounding, and reliable AI question answering. The paper concludes that RAG can support more context-aware and evidence-based answers, but its reliability depends on source quality, retrieval effectiveness, context selection, and answer verification.