An agentic codified framework for lower hallucination and better explainability in Legal question answering.

In this project, we aim to overcome hallucinations in LLM legal assistants by developing an agentic codified solution for legal question-answering (QA), which incorporates rule-based and (programming) code-based representations of law, as well as distributed reasoning and verification. The trustworthy legal research assistant will transform the daily search routine of legal professionals by speeding up the process and enhancing their efficiency. Through a conversational interface, a legal professional will be able to communicate with our legal assistant in natural language to build a full argument for a given case, which can itself be uploaded to the chat interface to add more context. The assistant answers each query, providing the cited sources which can also be consulted or summarized.

Keywords:ARTIFICIAL INTELLIGENCE (AI)

Disciplines:Natural language processing, Neural, evolutionary and fuzzy computation, Legal language and linguistics

Project type:Collaboration project