Using Fine-Tuning for Q/A model that can be trained on chat and mails available between user and agent

ChatGPT OpenAI allows us to fine-tune GPT-3.5 models. I want to make the dataSet that wI can use for fine-tune a model, I want to use data of real time chat between user and agent or communication happened via mail. I want to make that data in JSONL format and then fine tune a model so that model can also answer just like real time agent. I have tested and fine-tuned the model with my own dataset which I created manually but the problem is the fine-tuned model generates the answer randomly, not correct based on my custom dataset.

Is there any way to prepare dataset automatically and make the model only answer from my own fine-tuned dataset?

Any suggestion or ideas to solve this issue.