A Large Language Model based legal assistant for governance applications

Large Language Models (LLMs) have gained significant traction, primarily due to their potential disruptive influence across industries reliant on natural language processing. Governance stands out as one such sector. Notably, there has been a surge in research activity surrounding the implications of LLMs in deciphering complex legal corpora. This research offers substantial assistance to various stakeholders, including decision-makers, administrators, and citizens. This article focuses on the design and implementation of an LLM-based legal assistant tailored for interacting with legal resources. To achieve this, a real-world scenario has been chosen, incorporating models GPT3.5 and GPT4 as the LLMs, a well-defined legal corpus comprising European Union (EU) legislation and case law concerning the General Data Protection Regulation (GDPR), alongside a series of reference legal queries of varying complexity. Retrieval Augmented Generation (RAG) as well as agent methodologies are employed to seamlessly integrate the LLMs' functionalities with the customized dataset. The results appear to be promising, as the system managed to correctly address the majority of the legal queries, though with variable precision. Expectantly, the complexity of the queries severely impacted the quality of the outcome.