Large Language Models (LLMs) have significantly advanced human-computer interaction by enhancing semantic understanding and contextual awareness. However, they face challenges with domain-specific and cross-lingual queries due to their reliance on pre-trained knowledge. Retrieval-Augmented Generation (RAG) mitigates this limitation by integrating external knowledge retrieval, yet Dense Information Retrieval (DIR) methods often struggle with specialized domains and complex queries (CQ) requiring multi-document retrieval. To address these challenges, we propose an end-to-end framework that improves retrieval accuracy in cross-lingual settings. Our approach combines synthetic data generation for domain-specific, multilingual queries with fine-tuning of dense retrievers. We evaluate our framework across eight languages of varying resource availability, demonstrating enhanced retrieval accuracy, particularly in cross-lingual RAG scenarios. Key contributions include optimized domain-specific RAG, improved complex query handling, and robust cross-lingual adaptation.