Beyond SWE-Bench: Multi-language Automated Program Repair at MICAI 2025

Our paper “Beyond SWE-Bench: A Compiler-Assisted Pipeline for Multi-language Automated Program Repair” was published at MICAI 2025.

Research Summary

This work extends beyond the popular SWE-Bench benchmark to address automated program repair across multiple programming languages. Our compiler-assisted pipeline leverages:

• Static analysis for bug localization
• Compiler feedback loops for repair validation
• Multi-language support beyond Python-only benchmarks
• Improved repair accuracy through structured reasoning

Team

This research was a collaborative effort with:

• Moises Pineda
• Diego Luna
• Mariana Esquivel
• Juan Salazar
• Prof. Daniel Flores Araiza

The paper contributes to the growing field of AI-assisted software engineering by demonstrating how compiler integration can improve the reliability of automated repairs.

Originally shared on LinkedIn