Automating Judicial Efficiency with Transformer NLP and Deep Learning Models

The Indian judicial system is plagued by heavy delays caused by the increasing number of case files, long legal documents, and intricate transcripts. This study suggests a light, understandable AI architecture for enhancing the efficiency of the judiciary by simplifying legal document summarization, semantic retrieval of precedents, and evidence-based question answering. Our pipeline combines supervised extractive summarization via Logistic Regression and Linear SVM models, effective retrieval via BM25 and Doc2Vec embeddings, and a retrieval-first RAG (Retrieval-Augmented Generation) model for explainable query answering. The system is computation-efficient and maintains complete traceability of decision-making, thus being deployable in resource-limited judicial settings. Experimental testing on a collection of Indian verdicts illustrates competitive performance: the Logistic Regression model produces 66 % precision, 100 % recall, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$80 \% ~\mathrm{F} 1$</tex>-score, and 96 % accuracy on extractive summarization tasks, and the Linear SVM model depicts <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 0 \%}$</tex> precision, 100 % recall, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$66 \% ~\mathrm{F} 1$</tex>-score, and 92 % accuracy. These findings identify the framework as capable of providing interpretable, efficient, and responsible AI support within judicial proceedings, thereby decreasing cognitive load for legal practitioners and enhancing case understanding