"Design Twitter." Don't draw a single box yet.

transcript

The interviewer looks up. Design a Twitter-like platform. The clock starts. Draw boxes? Pick a database? Don't. Not yet.

You ask about scope first. Just the core feed, or DMs and search too? Core feed. Then scale. 100 million monthly active users. People view far more than they post. Assume a thousand reads per write. And latency. Feed loads under 200 milliseconds, at P99.

Now watch. 100 million users is where your estimates start. A thousand reads per write? Optimise the read path. Caching. Read replicas. And 200 milliseconds isn't a feature. It's a bar the architecture must meet.

Then say it all back. Add your own assumption: eventual consistency for the feed. And ask: does this sound right? Now they can correct you early.

That was step one. 3 to 5 minutes, clarifying. Step two, 5 to 7 minutes. Estimate traffic, storage, servers. Only now, step three. 15 to 20 minutes drawing the high-level design. Then 10 to 15 minutes deep in one or two parts.

You don't draw one box until you know what you're building. And how big it is. Ask first. Then draw.

Full breakdown in the lecture.