A Graph-Enhanced LLM-Based Question Answering System for the AI Act

Artificial Intelligence (AI) is increasingly integrated across sectors like healthcare, finance, and law, bringing both innovation and ethical concerns, such as bias, lack of transparency, and accountability. As AI systems grow in complexity, the need for robust governance becomes ever more critical. The EU AI Act addresses these challenges as the world's first comprehensive regulation on AI. It introduces risk-based classifications, transparency mandates, and governance measures to ensure alignment with ethical and legal standards. However, legal documents like the AI Act are often difficult to navigate due to their complex structure and language, complicating efficient information retrieval.

In this session, the speaker will present a graph-enhanced, LLM-based question-answering (QA) system designed to extract relevant information from the AI Act. The proposed solution combines: (i) a graph-based representation of the Act to improve document's organization and exploration, and (ii) a language model agent interacting with this graph to provide traceable and context-aware responses. Inspired by GraphReader and adapted for legal contexts, the model organizes the Act into Chunks, AtomicFacts, and KeyElements. Compared to traditional knowledge graphs or ontology-based legal QA systems, this approach offers greater flexibility and scalability, indicating improvements in the accuracy, relevance, and transparency of responses.

You will learn how combining LLMs with graph technology can improve the organization, interpretability, and accessibility of complex legal texts. The session will also demonstrate practical strategies for building scalable legal QA tools using graph databases and offer guidance on adapting this approach to other regulatory documents.