Contract review is essential to minimize risks during project execution. Missing critical clauses, particularly toxic clauses, can lead to increased construction costs, legal disputes, compensatory payments, and, in severe cases, project failure. However, this critical task still heavily relies on human expertise, leading to reduced review capabilities when experienced personnel leave. To address these issues, this paper proposes an automated contract review framework based on Retrieval-Augmented Generation (RAG) and open-source small Language Model (sLLM) technologies. The proposed framework utilizes a RAG pipeline tailored to contract data to efficiently retrieve semantically relevant information for each checklist category from contracts. Using dynamically generated prompts, the sLLM determines the presence of toxic clauses, suggests improvements, and provides a summary with the original sentence's location. Testing on actual contract documents showed that over 80% of necessary review items were successfully extracted, demonstrating the framework’s potential to effectively support contract specialists.

＊key words: Offshore Project, Contract Review Automation, RAG, sLLM