Building an AI-driven assistant for the Administrative and Legal Domain

Abstract

The rapid growth of digital information has created significant challenges in managing, organizing, and retrieving knowledge, particularly within legal and administrative domains. Conventional keyword-based search tools often fail to capture the contextual meaning of complex documents, leading to inefficiencies in decision-making and legal research. To address this issue, this study presents the design and implementation of an AI-driven assistant that integrates vector databases with large language models (LLMs) to support administrators and legal professionals. The proposed system leverages semantic embeddings to transform unstructured legal and regulatory texts into high-dimensional vector representations. Experimental evaluation was conducted using a legal dataset from the Libyan Academy for Telecom and Informatics (LATI), comprising more than 200,000 words of official laws and regulations. The system was tested with legal queries and assessed both automatically and through expert review. Results demonstrated that the assistant retrieved accurate, contextually relevant passages, significantly reducing response times compared to manual search. Legal experts confirmed that most answers were precise and practically useful, although further refinement is required for handling ambiguous or nuanced cases. The system improves administrative and legal efficiency by enabling semantic search beyond keyword matching and providing actionable insights for decision-makers. Future work will focus on expanding the legal corpus, refining query handling, and exploring advanced indexing techniques to improve scalability, accuracy, and adaptability across different domains.