RAG Based Legal Document Assistant with Verifiable Citations and Multilingual Support

Abstract

The increasing availability of legal information in digital form has created a need for intelligent systems that can provide reliable and understandable access to legal documents. Conventional language models may generate fluent responses but can produce unsupported information, particularly when answering complex legal questions or working with multilingual legal content. This paper proposes a Retrieval Augmented Generation based Legal Document Assistant designed to provide grounded responses with verifiable citations. The proposed framework combines multilingual query processing, semantic and lexical document retrieval, relevance based reranking, context aware answer generation, and citation verification within a unified architecture. User queries are first processed according to their language and transformed into representations suitable for multilingual retrieval. Relevant sections from a collection of legal documents are then retrieved and supplied to the language model as supporting context. Each generated response is associated with source references that allow users to trace the information back to the relevant legal document and section. A citation verification mechanism is introduced to identify unsupported statements and improve the reliability of generated answers. The proposed approach is evaluated using retrieval effectiveness, answer relevance, citation correctness, citation coverage, and groundedness across multiple languages. The framework aims to improve accessibility to legal information while reducing unsupported generation and making AI generated legal responses more transparent and traceable. The proposed architecture can serve as a foundation for multilingual legal information systems where accuracy, source verification, and responsible use of artificial intelligence are essential