Efficient retrieval of relevant legal documents is crucial in the legal domain, particularly for languages with limited computational resources such as Vietnamese. Vietnamese legal texts pose significant challenges due to their complex structure, formal language, and specialized terminology. Traditional retrieval systems often struggle to deliver accurate and relevant results in this context. This paper presents a fusion-based document retrieval system specifically designed for Vietnamese legal texts. By combining multiple retrieval methods, including traditional lexical matching and modern neural network-based techniques, we leverage their complementary strengths to enhance retrieval performance. Our system incorporates domain-specific knowledge and linguistic features unique to the Vietnamese language and legal texts, ensuring high relevance and accuracy in search results. We provide a comprehensive evaluation of our system, demonstrating its superiority over existing single-method retrieval approaches.