Shanghai Jiao Tong University has used domestic LLM MiniMax to "re-engineer" well-known Western simulation, image-processing and video-editing software — reading legacy software, then rewriting it with identical functions and an upgraded core, Shanghai Observer reported on November 1. The result, described as China's first LLM-native software reverse-engineering achievement, was presented at the World Laureates Forum's AI session in Lingang, Shanghai, which closed the same day.
CAD, CAE and CAM tools sit at the heart of aerospace, shipbuilding and automotive R&D and have long been dominated by Western vendors whose value rests on inaccessible source code. Traditional reverse engineering — experts disassembling binary software by hand — takes years and rarely recovers the design rationale inside.

SJTU's method, "LLM-based software re-engineering," runs in three steps. First, modify the model itself to gain native deep semantic analysis of core software logic, moving from recognizing code to understanding intent. Second, generative reconstruction: having grasped the original design, the model rewrites the code while trimming redundancy, improving speed and patching vulnerabilities. Third, cross-checking by independent AI agents that verify logic, diff old and new versions and hunt for flaws.
The project is led by Kong Linghe, IEEE Fellow and distinguished professor at SJTU. The team validated the approach on multiple text models at home and abroad and landed on MiniMax for its strengths in reverse analysis and logic recovery. In one case, a US simulation package of 30,000 functions has had 5,000 core functions re-engineered with aligned functionality and comparable performance; these, with code, documentation and test cases, are to be fully open-sourced.
Context: a tool called REA ("Reverse Engineer Anything") has collected hundreds of thousands of GitHub stars, as hobbyists strip smart-TV boot ads or link PCs to smart toilets. Industrial software is the deep end of reverse engineering — and LLMs compress months of expert work into a systemized pipeline.
For China's industrial-software effort, re-engineering offers a path around ground-up development: not copying code, but rewriting a better implementation after understanding the mechanics. Whether the 5,000 open-sourced functions grow into a reusable toolchain will decide how far this road goes.