I built the project because we often don't know how the AI works and not everyone knows how to fine-tuning the model. Now we can see what each layer of the model actually does and we can make fine-tuning simpler. We can collect training data while we chat. If an answer is not good, fix it and save the version as a training example.