Whenever you're ready, walk me through how you'd approach this.
If a model performs very badly on training data, what could be the reason, and how would you fix it?
