Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The exponential growth of digital legal documentation has created unprecedented challenges for automated information extraction in enterprise environments. This article presents a comprehensive evaluation of generative AI-assisted regular expression synthesis for extracting key contractual clauses from legal agreements. The framework employs Claude-3 Sonnet to automatically generate extraction patterns, comparing performance against traditional human-authored alternatives across technology, healthcare, and finance sectors. Evaluation encompasses precision, recall, F1-scores, development time, and syntactic complexity metrics using a corpus of anonymized enterprise contracts. Results demonstrate substantial efficiency improvements in pattern development while maintaining superior accuracy performance compared to manual approaches. The implementation incorporates sophisticated prompt engineering strategies, active learning feedback loops, and multi-layered validation frameworks to ensure reliability and prevent over-generalization. Domain-specific adaptations reveal sector-dependent performance variations, with technology contracts yielding optimal results and healthcare agreements presenting greater complexity challenges. Comprehensive safeguards, including confidence scoring mechanisms, human oversight integration, and regulatory compliance measures, address critical concerns regarding bias mitigation and systematic validation. The automated synthesis approach exhibits reduced pattern complexity while achieving enhanced maintainability across diverse contractual environments, supporting broader adoption of AI-assisted legal document processing technologies in enterprise settings.