For each variable:

1. It asks the LLM to write a description of what the variable does

2. It asks for a good variable name based on the description from 1.

3. It uses a custom Babel plugin to do a scope-aware rename

This way the LLM only decides the name, but the actual renaming is done with traditional and reliable tools.