This project solves the legal outcome prediction challenge by creating an Optimized Hierarchical Attention Network (HAN) that can examine legal documents and forecast case judgments. Legal experts tend to face laborious manual reading of lengthy texts. To counter this, the model uses multi-level attention mechanisms that concentrate on key information at word, sentence, and paragraph levels, increasing contextual comprehension across different case types and jurisdictions. The main objective is to design and train an optimized HAN model for the identification and prediction of legal outcomes with adaptability to different case types and jurisdictions. The optimized HAN has a refined architecture with layer-wise attention refinement, sophisticated preprocessing, and domain-specific embeddings such as LegalBERT and T5. LegalBERT extracts legal vocabulary and context, and T5 enhances text-to-text transformations for making coherent predictions. These optimizations provide strong performance and flexibility across different legal domains. The system provides real-time predictions through an easy-to-use interface, assisting legal professionals in making effective decisions, minimizing manual labor, and ensuring consistency in legal outcomes.