SENTINEL: Multimodal AI Framework for Contract Risk Analysis and Negotiation Strategy Generation

Abstract: Manual examination of contractual documents demands extensive human effort and often leads to inconsistent identification of risk-bearing clauses. This work introduces SENTINEL, a multimodal analytical framework designed to automate risk assessment and generate context-aware negotiation recommendations from legal agreements. The system integrates a hybrid OCR pipeline combining CRNN and rule-based recognition for text extraction, followed by clause segmentation and domain-adapted LegalBERT classification for risk prediction. A retrieval-augmented mechanism further enhances the system by leveraging semantically similar historical clauses to generate negotiation suggestions using large language models. Evaluation conducted on 847 real-world contracts demonstrates 87.3% classification accuracy and strong agreement with expert assessments (r = 0.81). The results indicate that combining structured document understanding with retrieval-driven generation significantly improves efficiency and supports informed decision-making in legal workflows.

Keywords: Contract Analysis, Legal AI, Risk Classification, Retrieval-Augmented Generation, Natural Language Processing, LegalBERT, Document Intelligence