Gracenote.ai: Legal Generative AI for Regulatory Compliance

Abstract

We investigate the transformative potential of large language models (LLMs) in the legal and regulatory compliance domain by developing advanced generative AI solutions, including a horizon scanning tool, an obligations generation tool, and an LLM-based expert system. Our approach combines the LangChain framework, OpenAI’s GPT-4, text embeddings, and prompt engineering techniques to effectively reduce hallucinations and generate reliable and accurate domain specific outputs. A human-in-the-loop control mechanism is used as a final backstop to ensure accuracy and mitigate risk.

Our findings emphasise the role of LLMs as foundation engines in specialist tools and lay the groundwork for building the next generation of legal and compliance applications. Future research will focus on extending support across multiple jurisdictions and languages, refining prompts and text embedding datasets for enhanced legal reasoning capabilities, and developing autonomous AI agents and robust LLM-based expert systems.

Keywords: AutoGPT, compliance, expert systems, GPT-4, GRC, LangChain, large language models, legal generative AI, legal text embeddings, prompt engineering, regulation