Structuring Large AI Projects and Repositories

The author maintains a large codebase of 15 C++ apps across 4 repos, with a common repo used as a submodule. The codebase is too large for LLM tools like Claude Code to effectively make changes without breaking other parts. The author plans to refactor and rewrite the codebase to make it more AI-friendly.

Details

The author maintains a total of 15 C++ apps distributed across 4 top-level repos, including 1 general-purpose "common" repo (git submodule) where they store code required by more than 1 repo. The codebase is organized into 3 clusters of 5 related embedded apps, spanning from backend to frontend to monitoring. Each cluster is around 350k LOC, and the main backend app is 120k LOC. The author plans to refactor and rewrite the codebase, which they estimate could be cut in half by turning a lot of it into parameterizable shared code. However, even cut in half, the codebase is still too large for LLM tools like Claude Code to effectively make changes without breaking other parts of the system.