<p>Legal decision prediction is a judicial assistance system that recommends the decision components, such as applicable statutes, prison, and penalty, by examining the given case document, named fact description. The worldwide judicial system is facing a pressing need for technical assistance that addresses the mounting backlog of cases across diverse courts. This paper explores deep learning techniques to create a decision support system that enhances the technical support for the judicial system by integrating artificial intelligence and ambient technologies. Most existing works on the judicial system did not adequately concentrate on the semantics embedded in the fact description that impact the decision. The proposed framework provides the advantages of pre-trained transformers for understanding complex, unstructured, lengthy legal documents. The model draws the in-depth semantics of the given fact description at multiple levels, i.e., chunk and case document level, by following the divide and conquer approach. It creates a concise view of the given fact description using the multi-stage semantic learning framework as per the original court case document structure, then predicts the judgment. We tested the model’s performance on two Indian datasets and got promising results. The model achieved a 5.97% improvement on the Indian legal documents corpus dataset and reduced performance degradation with extended training epochs compared to existing approaches. It significantly decreased the code execution time, resulting in lower consumption of system resources like the GPU and memory. The model outperformed several baselines, with an improvement of 10.61% over the Indian legal statute identification dataset.</p>