ReverseAI: Large-Scale Compute Framework for AI-Driven Reverse Execution and Dynamic Program Analysis

This project aims to leverage EuroHPC JU compute resources to advance AI-assisted debugging and self-correcting software systems through large-scale training and benchmarking of neural architectures for reverse execution and dynamic slicing. Building upon the “Bug Locate & Fix” and “AI-SysDev Fine-Tuning” components developed within the AI4SWEng Horizon EU Project, the project will train and evaluate models that simulate reversible program execution in multi-threaded and multi-processor environments. These models will use deep learning architectures inspired by BugLab and self-supervised code reasoning frameworks to identify, localise, and automatically repair software bugs without labeled data.

The compute resources will enable large-scale experiments on self-supervised model training, pseudo-code generation, reverse execution simulation, and dynamic slicing performance analysis using millions of real-world code samples. The expected outcomes include high-efficiency, interpretable debugging agents capable of rewinding or replaying execution scenarios, predicting fault origins, and autonomously proposing fixes. The work contributes to developing next-generation tools for energy-efficient, trustworthy, and self-healing software systems, directly supporting the AI4SWEng and related AI Factory initiatives in AI-assisted software engineering.