Trustworthy Contracts AI: Explainable Multi-Label NLP with Legal-BERT, T5, and SHAP

Legal contract analysis remains a time-intensive manual process despite advances in natural language processing. This paper presents an end-to-end explainable AI framework for contract analytics that combines multi-label clause detection with Legal-BERT, abstractive summarization via T5, and layered interpretability using SHAP, LIME, and attention visualization to generate lawyer-readable rationales. Trained and evaluated on the Contract Understanding Atticus Dataset (CUAD) comprising 510 contracts across 41 clause types, the system addresses long-document constraints and severe class imbalance through sliding-window inference and weighted losses. Experimental results demonstrate strong multi-label performance with F1-micro of 0.8924 and low Hamming loss of 0.0023, alongside competitive ROUGE scores for summarization (ROUGE-1: 0.6054). Beyond offline metrics, a containerized web application demonstrates practical deployment viability. The framework's transparent attributions over legally salient spans significantly improve practitioner trust and decision support in high-stakes legal NLP applications, advancing responsible AI adoption in legal technology.