I have a legacy codebase of size approximately 2 GB written primarily in C++. There are lot of inhouse libraries that are build on top of primitve C++. For example we have our own implementation of vectors, queue, maps etc. Also libraries built around sockets to provide HTTP functionalities etc. Any new application to be written has to use methods provided by these inhouse libraries.
My question is, Is there any way AI/ML can be helpful here to train LLM models for these inhouse libraries written in C++ ( or any language in general) so that application function codes can be auto generated using methods provided in these legacy inhouse libraries? I don’t want to use ChatGPT for security reasons and also it will provide me code with Standard C++ which can’t be used directly in our system. Any suggestions on how to train LLM with legacy code base will be really helpful.

Tried to search existing options but could not find any satisfactory answers.