Show HN: LLM-assisted reconstruction of partially decompiled Minecraft 26.1.2 (github.com)

A new research project, "demcstify", has launched aimed at reconstructing the game Minecraft version 26.1.2 from its partially decompiled sources into fully operational client and server binaries, using Large Language Models (LLMs). This ambitious initiative serves as an experimental framework that coordinates LLM agents against a deterministic bytecode oracle to ensure that the reconstructed artifacts remain entirely local and do not contain any copyrighted Minecraft files. The project is compliant with Mojang’s and Microsoft’s rights, emphasizing that it is a reconstruction lab rather than a traditional code repository. This development is significant for the AI and machine learning community as it pushes the boundaries of automated source reconstruction, demonstrating that LLMs can help address the inherent inaccuracies found in decompiled code. By treating original class files as the authority and employing a structured pipeline where decompilation, source repair, and bytecode comparison are meticulously managed, "demcstify" aims for a level of fidelity not previously achievable. The use of a multi-agent system allows for concurrent work, enhancing efficiency and showcasing how AI can be effectively integrated into software reconstruction and reverse engineering tasks. This approach could inspire future endeavors focused on improving code analysis and development workflows in complex software environments.