Problem is we always translate the whole file. Suggestions for implementation is we create a mapping that is cached and we detect what line changed, and submit that to the AI services.