While Retrieval-Augmented Generation (RAG) has shown impressive results across various applications, including question answering, summarization, and dialogue generation, it also has limitations, especially when it comes to personalized answers. One significant challenge is that RAG models typically rely on generalized retrieval methods and do not always tailor responses to individual user profiles or preferences. This can result in answers that lack personalization, failing to account for the user’s specific context, needs, or past interactions. Therefore, this research introduces a new approach to improve the performance of existing RAG-based systems by incorporating three main features: question rewriter, multi-embedding, and user profile information. Experimental results on three datasets—HotPotQA, SQuAD 1.1, and the Vietnamese AI Vietnam dataset—show that the proposed system improves the performance of existing RAG-based systems in terms of generating personalized answers.