Advancing legal accessibility in Bangladesh through AI-powered assistance and natural language interfaces

Abstract

In a country where laws and regulations are as difficult to understand as the cases they govern, the path to justice is anything but straightforward. For many in Bangladesh, the legal system’s complexity acts as a gatekeeper because most of the people can not simply comprehend its complex language. Our research presents a Retrieval Augmented Generation (RAG) system that integrates large transformerbased language models like Mistral-7B, deCILM-7B, LLaMA 3.1-8B, and others with a specialized legal retrieval module containing Bangladeshi laws and regulations. This allows the system to interpret user inquiries using natural language, retrieve relevant legal information from the corpus, and generate detailed responses grounded in and applying the country’s legal code - complete with applicable citations. At the heart of it all is an embedded Retrieval Augmented Generative system that combines neural automation and strong encoder-decoder transformers for generating legal documents and classifications quickly, ensuring legal accuracy. And the human-in-the-loop verification process to review and modify output allows users to get even more accurate results. This can massively improve general people’s understanding of legal matters as well as it can work as a lawyer’s helping hand too.