NIC has developed a RAG-based judicial search assistant for Supreme Court justices. It will serve as a specialized search service to help judges quickly find specific information to support the delivery of justice.
To create this assistant, Supreme Court judgments have been converted into embeddings with the NVIDIA NeMo Embedding microservice and stored in a GPU-accelerated vector database. In response to verdict-specific queries, the RAG solution retrieves information from the vector database using the NVIDIA NeMo Retriever microservice. Results are refined with the NeMo Reranking microservice and fed into a TensorRT-LLM-optimized large language model deployed using LLM NIM, which utilizes the query, prompt, and retrieved context to generate a natural response containing the necessary information and associated verdict documents. The solution was deployed on DGX A100 Tensor Core GPUs and uses NVIDIA AI Enterprise software for efficient embedding generation, reranking, and inference.
The solution will be available to all of India’s Supreme Court judges with plans to scale the judicial assistant to the entire judicial system.
NIC has also developed abstractive and extractive document summarization to generate summaries of pleadings. It's crucial for those needing a quick, accurate summary, as manual summarization is time consuming.