Automated Bug Detection And Fixing Using T5-Small Transformer Model: A Multi-Language Approach

Software bugs remain one of the most persistent challenges in software development, consuming 50-75% of developer time and costing the global economy over 2 trillion annually. This paper presents a multi-language approach to automated bug detection and fixing using the T5-Small transformer model. We construct a dataset of 2, 600 real bug examples from Defects4J, BugSwarm, QuixBugs, GitBugs, and 500 novel multi-error examples. The T5-Small model (60M parameters) is fine-tuned with optimal hyperparameters. Our evaluation framework employs seven metrics with mathematical formulations. Experimental results demonstrate 68. 46% Normalized Exact Match, 93. 74% F1 Score, and 99. 55% ROUGE-1. The model performs effectively on both Python (70. 0%) and Java (65. 0%). All artifacts are released open-source.