How I Use AI to Prep for System Design Interviews

System design interviews are broken#

You walk in. They ask you to "design Twitter" in 45 minutes. You draw boxes on a whiteboard while someone judges whether your arrows point in the right direction.

It's a terrible way to evaluate engineering talent. But it's also not going away. So let's talk about how to actually get good at it.

The old way: books and YouTube#

Everyone says "read Designing Data-Intensive Applications." It's a great book. I've read it twice. But it doesn't help you practice. You can't interact with it. You can't ask "what happens if this database goes down?" and see the cascading failure.

YouTube system design videos are better — you watch someone think through the problem. But you're a passive observer. You don't build the muscle memory of making architectural decisions yourself.

The new way: interactive simulation#

Here's what changed my prep completely: instead of reading about how Netflix works, I describe Netflix's architecture and watch it come alive. Now I can actually interact with it. Click the CDN node — what happens if it goes down? Run a stress test on the transcoding service. How many replicas do we need?

This is the difference between reading about swimming and actually getting in the pool.

My interview prep process#

Step 1: Generate the base architecture#

I type the system I need to design: "Design a ride-sharing app like Uber." In 10 seconds I have a full architecture with all the major components.

Step 2: Drill into each component#

I click on the matching service. What's inside? How does it handle location updates? What database does it use? Each drill-down gives me the sub-architecture.

Step 3: Run the audits#

For each component, I run:

• Stress test — where are the bottlenecks?
• Security audit — what are the attack vectors?
• Cost estimate — what's this going to cost at scale?

These are exactly the questions interviewers ask. Now I have practiced answering them.

Step 4: Compare approaches#

The interviewer says "would you use SQL or NoSQL here?" Instead of guessing, I use Compare Mode to generate both approaches side-by-side. I can articulate the trade-offs because I've seen both architectures.

Step 5: Export as talking points#

I export the architecture as an AI prompt and paste it into Claude for a mock interview. "Here's my architecture for a ride-sharing app. Ask me tough questions about it."

The systems I practice with#

Here are the top 10 systems that come up in interviews. I've practiced all of them:

1. URL shortener (Bitly)
2. Chat application (WhatsApp/Discord)
3. Social media feed (Twitter/Instagram)
4. Ride-sharing (Uber/Lyft)
5. Video streaming (Netflix/YouTube)
6. E-commerce (Amazon)
7. Search engine (Google)
8. File storage (Dropbox/Google Drive)
9. Notification system (push/email/SMS)
10. Rate limiter / API gateway

For each one, I generate the architecture, drill into components, run audits, and practice explaining my decisions out loud.

It actually works#

I went from bombing system design interviews to getting offers at two FAANG-adjacent companies. The difference wasn't reading more — it was practicing interactively.

When the interviewer asked "what happens when your database goes down?", I didn't freeze. I'd already simulated that exact failure mode. I knew the answer because I'd seen it play out.

Start now#

Your next system design interview is coming. Don't just read about architecture — practice building it. Describe a system, watch it come alive, break things, fix things. That's how you learn.