How should straucter my dataset for llm?

well i want to create gpt bot to help me remmber what said in video, i have data like this

video url : 
video text : { time:1:1,text:blahblah } 

i want to fine tune gpt to find text and complete sentence with time and video url.

all example/sample i saw so far have big data with no label on it(??)

Repl link:

code snippet