Justice Lens: An AI-Powered Legal Assistant for Simplifying Indian Law

Abstract

Obtaining legal information in India is a major barrier to entry for the common individual due to the statutory language being dense and the exorbitant prices associated with hiring legal experts. Although conversational LLMs are a great starting point in helping users find answers, they often tend to hallucinate cases and fail to cite credible sources. In this work, we introduce Justice Lens, a legal expert powered by AI. The framework is based on the Retrieval-Augmented Generation paradigm, which allows it to provide accurate and reliable legal advice. Moving away from keyword-driven search to semantic search through the use of high dimensional vector embeddings, the architecture enables instant and intent-based queries. Our RAG model makes use of a Pinecone vector store with Approx- imate Nearest Neighbors (O(log V ) retrieval time) along with the Google Gemini LLM to convert complex Indian laws into simple English. We demonstrate through our experiments that our framework is successful in avoiding any AI hallucination by ensuring a ”duty to verify” constraint in every output.

Keywords: Artificial Intelligence, Legal Technology (LawTech), Semantic Search, Vector Embeddings, Transformer Models, Indian Judicial System, Explainable AI (XAI).