I am using Gemini to analyze a vast amount of feedback data from customers. I want to ask the model questions like “what are the most common complaints?” and “what do customers like about our services?”.

I am currently just prefixing the prompt with the relevant data but this is slow and often throws errors when the data is too large.

Fine tuning sounds like the way to go but I only have the raw data, not a list of inputs/outputs for the model to train on. Is there a way to fine tune the model with just all the data? Or is there another way to change the state of the model to allow for faster prompt responses?

Note that I have also tried RAG architecture, but this does not seem very useful in that I am already providing a subset of the data (and do not need semantic querying on the data).