abstract
- Automated program repair (APR) research predominantly focuses on Python environments, creating significant infrastructure gaps for compiled languages like C, C++, and Java that dominate production systems. We present the first systematic pipeline addressing multi-language APR infrastructure limitations through compiler-assisted dataset curation and paradigm-aware evaluation frameworks. Our approach combines a DFA-based code classification system achieving 92.4% accuracy in programming paradigm detection with systematic dataset filtering that processes over 3 million samples to extract 30,000 high-quality object-oriented examples. Initial evaluation on Qwen3-14B using LoRA fine-tuning reveals critical adaptation thresholds: effective multi-language adaptation requires modification of approximately 1.2% or more model parameters, with lighter fine-tuning underperforming baseline models. Our open-source pipeline provides end-to-end infrastructure from compiler-assisted dataset curation to cloud deployment, enabling systematic research advancement in multi-language automated program repair and establishing methodological foundations for compiler-assisted machine learning across diverse programming environments. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.