Can a LLM master a subject through training and fine-tuning, and then generate questions?

[zgcatc]

I want to use a large language model to create a personalized learning system for a specific subject. The large language model will generate questions, identify knowledge gaps based on incorrect answers, and reinforce learning by generating questions targeting those knowledge gaps. I want to know if a large language model can master the knowledge of a subject through fine-tuning and then generate various types of questions. Is this financially feasible, considering we plan to commercialize it? Or do we need to train from scratch? Can we implement this using open-source packages?