Evaluating and Improving the Quality of LLM-Generated Code
Glaucia Melo, Jessica Pourleyli, Genevieve Caumartin, Corey Yang-Smith, Diego Costa, and Ahmad Abdellatif
Date:
Three-hour conference tutorial at 34th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (FSE 2026), Montréal, Canada
Co-presented with Glaucia Melo, Jessica Pourleyli, Genevieve Caumartin, Diego Costa, and Ahmad Abdellatif at FSE 2026 in Montréal, Canada (July 5, 2026).
The tutorial introduced a reusable end-to-end pipeline for evaluating the maintainability, reliability, and security of LLM-generated code. Topics included static analysis with tools such as SonarQube and Bandit, comparison against model/prompt/human baselines, agentic remediation, explainability for trust calibration, and bias-aware evaluation.
An archival companion-proceedings summary is available (DOI: 10.1145/3803437.3804900).
Recommended citation: Glaucia Melo, Jessica Pourleyli, Genevieve Caumartin, Corey Yang-Smith, Diego Costa, and Ahmad Abdellatif. "Evaluating and Improving the Quality of LLM-Generated Code." Three-hour conference tutorial, 34th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (FSE 2026), Montréal, Canada, July 5, 2026. Companion paper DOI: 10.1145/3803437.3804900.
Download Paper | Download Slides | Download Bibtex