Ermanno Francesco Sannini

My research activity operates at the intersection of Software Engineering, Blockchain Technology, and Artificial Intelligence, with a primary focus on improving the quality, security, maintainability, and efficiency of Ethereum smart contracts written in Solidity. Driven by empirical software engineering methodologies, static code analysis, and Natural Language Processing (NLP), my work addresses the unique technical and operational challenges imposed by the immutable, high-stakes execution environment of decentralized applications.

A major pillar of my research explores semantic code reuse and documentation reliability in smart contracts. Through large-scale empirical studies, I investigated Type-4 (semantic) clones in Solidity functions by combining syntactic similarity metrics with natural language comment semantics (utilizing SBERT, BERT, and Large Language Model proxies). This work established empirical taxonomies demonstrating how semantic code variations represent deliberate engineering trade-offs focused on gas optimization, modularization, and security enforcement. Complementing code clone detection, I systematically analyzed documentation quality and Code-Comment Inconsistencies (CCIs) across high-usage and low-usage smart contract projects. Through this, I developed a domain-specific taxonomy—spanning Business Logic, Security, Interface, and Delegation—revealing how misleading documentation and comment drift introduce cognitive blind spots, mask severe security vulnerabilities, and cause unexpected operational failures.

In parallel, my research investigates pre-deployment resource optimization and open-source maintenance dynamics. To help developers identify execution bottlenecks early in the development lifecycle, I designed machine learning models (such as Random Forest and Gradient Boosting) trained on static code metrics and embeddings to accurately predict gas-intensive functions before contract deployment. Furthermore, I identified and statistically validated six recurring architectural design patterns—such as storage access in modifiers, user role lists, and complex withdrawal logic—that significantly drive transaction costs on the EVM. Finally, my work examines post-deployment maintenance by mining open-source collaborative repositories (e.g., GitHub), characterizing issue-tracking behaviors, resolution times, feature enhancement workflows, and collaborative developer communications in blockchain-based projects.