Seeking Ideas for Training a Model to Assist with Writing ETL Rules in JSON

Hey everyone,

I’m working on a project that helps with writing ETL rules in JSON format, which are used internally at my company.
The goal is to train a language model to understand ETL requirements and generate accurate JSON rules based on user input. I have the internal documentation for the rules with me.

Can you please guide me how can I go about doing this? (Dataset Format? Model? Fine-tuning/RAG?)