Scholarlink

<p>This paper introduces “LegalEase,” an advanced legal question-answering (LQA) system specifically designed to simplify access to Indian legal knowledge. Built on the Retrieval-Augmented Generation (RAG) framework, the system integrates key technologies, including FAISS for efficient indexing, ChromaDB for semantic retrieval, and a multilingual translation feature. The system processes user queries by retrieving relevant legal texts, encoding them into embeddings, and using a large language model (LLM) to generate precise answers grounded in Indian legal frameworks like the Bhartiya Nyaya Sanhita (BNS) and the Indian Penal Code (IPC). A core objective of this research is to bridge the gap between complex legal language and user comprehension, thereby providing an accessible platform to enhance legal literacy in India. The study also evaluates the performance of three different LLMs—GPT-3.5-turbo-instruct, Gemini-2.5-flash, and Llama-3.1—on their ability to generate accurate and relevant legal answers for LegalEase. The evaluation uses a multi-faceted framework with metrics like ROUGE, MPNet score, and RAG-specific measures. Findings indicate that while each model has unique strengths and weaknesses, Gemini-2.5-flash performed best on ROUGE scores, while GPT-3.5-turbo-instruct excelled in semantic similarity based on the MPNet score. This research offers valuable insights into the practical application of LLMs for legal LQA systems and provides guidance on selecting the most effective models for such purposes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Redefining legal access: a RAG-based AI system for Indian law

  • 期刊论文

  • 发表日期：

    2025年10月11日

查看全文
  • Faiz Mubeen,
  • Aasar Mehdi,
  • Md. Asraful Haque,
  • M. Z. M. Nomani,
  • Nawab Shahzeb Uddin

摘要

This paper introduces “LegalEase,” an advanced legal question-answering (LQA) system specifically designed to simplify access to Indian legal knowledge. Built on the Retrieval-Augmented Generation (RAG) framework, the system integrates key technologies, including FAISS for efficient indexing, ChromaDB for semantic retrieval, and a multilingual translation feature. The system processes user queries by retrieving relevant legal texts, encoding them into embeddings, and using a large language model (LLM) to generate precise answers grounded in Indian legal frameworks like the Bhartiya Nyaya Sanhita (BNS) and the Indian Penal Code (IPC). A core objective of this research is to bridge the gap between complex legal language and user comprehension, thereby providing an accessible platform to enhance legal literacy in India. The study also evaluates the performance of three different LLMs—GPT-3.5-turbo-instruct, Gemini-2.5-flash, and Llama-3.1—on their ability to generate accurate and relevant legal answers for LegalEase. The evaluation uses a multi-faceted framework with metrics like ROUGE, MPNet score, and RAG-specific measures. Findings indicate that while each model has unique strengths and weaknesses, Gemini-2.5-flash performed best on ROUGE scores, while GPT-3.5-turbo-instruct excelled in semantic similarity based on the MPNet score. This research offers valuable insights into the practical application of LLMs for legal LQA systems and provides guidance on selecting the most effective models for such purposes.