An LLM-Powered Agentic Framework for Automatically Converting Construction Contracts into Smart Contracts

Abstract

Blockchain‑enabled smart contracts have recently gained significant attention for their ability to self‑execute predefined terms when specific conditions are met, offering great potential to advance construction automation. However, practical adoption in the construction industry remains limited due to the difficulty of automatically generating reliable smart contracts that ensure both safe deployment and compliance with construction-specific legal requirements. This study proposes an LLM‑powered “contract‑to‑smart contract” automation framework in which multiple AI agents collaboratively generate smart contracts that align with all legal and contractual requirements. We designed chain‑of‑thought (CoT) prompts to guide initial smart contract generation, then integrated a Solidity compiler and Slither security checks to ensure deployability and security. Detected errors are fed back as prompts for iterative self‑refinement. Lastly, a graph‑based process state analysis module was integrated to verify that the resulting smart contracts meet all behavioral requirements of the source contracts. We piloted the framework using two ConsensusDocs clauses. The results indicate high accuracy in complying with contractual requirements when present in the smart contracts and highlight areas for further improvement in completeness. This study contributes a practical step toward adopting smart contracts in the construction industry.