NLPALA: An NLP based artificial legal assistant leveraging retrieval-augmented generation in coordination with large language models

Abstract

In recent times, many noteworthy explorations have been conducted in the spectrum of legal affairs worldwide, but the legal sector in Bangladesh has yet to showcase prominent improvement. This research addresses the challenges in Bangladesh's legal affairs and proposes solutions using Natural Language Processing (NLP) techniques integrated with the latest trends in Large Language Models (LLMs). NLPALA represents a groundbreaking innovation in legal aid, offering comprehensive support to users in areas related to women's and children's prevailing acts. In our experiments, we explored diverse LLM models like GPT-3.5 Turbo, GPT-4, Claude-3.5 Sonnet and Gemini Pro, discovering that Claude-3.5 Sonnet performed notably better. By integrating this model with our proprietary dataset, we observed that using RAG with Claude-3.5 Sonnet significantly accelerated response times for user queries compared to other models. From answerings basic queries to assisting with complex legal procedures, NLPALA empowers users with accurate knowledge. Our research focuses on the development of NLPALA to use its latest methods to provide a good experience for individuals in Bangladesh.