Hybrid NLP Framework for Contract Risk Assessment-A Dual-Agent Approach Combining Roberta and Rule-Based Analysis with Unified Decision-Making

Efficient contract analysis in the legal domain demands accuracy, traceability, and interpretability, yet existing NLP systems often rely on opaque neural models, limiting transparency critical for law and compliance. We propose a dual-agent NLP frame-work that integrates high-performance contract analysis with explainability. A Clause Extraction Agent employs fine-tuned RoBERTa ensembles with pattern matching and semantic validation to identify key clauses, while an Enhanced Risk Assess-ment Agent combines rule-based legal logic with LLM-driven insights for vagueness detection, contextual risk evaluation, and clause rewriting recommendations. Modular design enables interpretability, focused debugging, and avoids black-box limitations. A proof-of-concept demonstrates significant improvements in clause extraction and risk assessment over baseline methods. Stateless data handoffs ensure full transparency from extraction to actionable recommendations. A Streamlit interface provides inter-active dashboards for overview, detailed clause analysis, risk assessment, and rewrite guidance. This framework establishes a blueprint for explainable, LLM-integrated legal AI suitable for real-world deployment.