We use a mixture of LLMs (e.g. to identify the important variables to trace), language servers (to understand where a variable is originally defined and where it's used).