Turn vague requirements into the few numbers that bound a design before you draw a single box.
You are not building a model. You are finding the single estimate that decides whether this fits on one node or needs a fleet. Peak, not average, and order-of-magnitude is enough.
Like estimating how many guests are coming before you decide how much food to cook. You don't need the exact headcount, just "about 10" vs "about 500", because that changes everything about the plan.
Requirements become QPS, storage, bandwidth, memory and a latency budget. Find the one that bounds the design and state your assumptions as you go.