NyayaSetu: An Explainable AI-Powered Legal Advisory and Predictive Analytics Platform for the Indian Legal Ecosystem

The Indian judicial system poses problems owing tothe sheer number of cases pending more than 43 million and thecomplex nature of accessing the judiciary process. The purpose ofthe current paper is to analyze a one-stop legal advisory solutioncalled “NyayaSetu” or “Bridge to Justice”, which is powered byArtificial Intelligence.Apart from the RAG pipeline using ChromaDB and Llama3.3 70B for zero hallucinations conversational legal advice, wewill also employ the XGBoost classifier for aiding us in ourdata processing task. The prediction task involves building aclassification model on an Extreme Gradient Boosting classifier.Training was done on self-created data comprising 1,400 casesdistributed among six domains of law. There are 27 features inthe dataset. In solving the black box problem, SHapley AdditiveexPlanation (SHAP) values were computed per case and suppliedto the user.The 5 fold cross validation value is 97.71% (±1.10%). Con-versely, the test score has an accuracy of 97.50%. The F1-scorefor each class is shown below: Favourable – 0.990 Unfavourable– 0.978 Settled – 0.909Index Terms—Legal AI, Retrieval-Augmented Generation, XG-Boost, SHAP, Indian Law, Predictive Analytics, NLP