BNS Mitra: RAG-Optimized LLM Based AI-Powered Legal Virtual Assistant

This paper presents BNS Mitra, a Virtual legal assistant for Bharatiya Nyaya Sanhita (BNS), which is powered by Artificial intelligence and is designed with RAG (Retrieval Augmented Generation) framework. Since many stakeholders exist: law enforcement, advocates, judiciary members and common people, BNS Mitra assisted by Meta's LLAMA 2 model is helpful in translating user-described incidents into legal sentences to determine the most relevant sections of BNS. BNS Mitra incorporated FAISS (Facebook AI Similarity Search) that provides the encoded BNS database for encoding vectors to use as a database suitable for user queries, thus allowing contextual legal advice to be provided in due time. Due to its simple design, the general public is able to obtain legal information. On the other hand, the time spent carrying out manual research is reduced since professionals are able to quickly obtain relevant sections. In tests BNS Mitra managed to achieve 87% accuracy regarding which BNS sections were most closely related to user input. The software performed well enough in simple cases, but some issues were identified in more complex cases. This paper presents the capacity of AI as a tool to diminish the knowledge gap in the Indian legal system and recommends the idea of expanding the scope of BNS Mitra to other legal areas/fields and language models.