Hi!
It sounds like you were using fine-tuning to inject specific knowledge. Unfortunately, this is not what fine-tuning (under the OpenAI fine-tuning endpoint) is intended for.
You should look into Retrieval-Augmented Generation (RAG) practices instead.
Here’s an OpenAI guide that puts the different pieces together in terms of when to use fine-tuning vs. RAG as well as explains how RAG works: https://platform.openai.com/docs/guides/optimizing-llm-accuracy/understanding-the-tools
Let us know if you have specific follow-up questions.