System design interviews and AI copilots

What a system design round tests

In a system design round, the interviewer gives you an open problem such as "design a URL shortener" or "design a chat app" and watches how you think. There is no single right answer. They want to see you ask good questions, make sensible estimates, choose parts and explain the trade-offs, all while drawing a clear diagram.

Most people struggle with pacing, not knowledge. They spend twenty minutes on requirements, or jump to a database choice before they know the load. A fixed path through the round fixes most of that.

A six-step path through the round

1. Requirements. Ask who uses it, what it must do, and what matters most: speed, cost or never losing data. Write the answers on the board.
2. Estimates. Rough numbers: users, requests per second, storage per year. Round hard; the point is the order of magnitude.
3. API and data. A few key endpoints and the main tables or records.
4. High-level design. Boxes and arrows: clients, load balancer, services, database, cache, queue. Walk one request through it.
5. Deep dives. Pick the hardest part and go deep: scaling the database, caching, ordering, failure handling.
6. Wrap up. Bottlenecks, what you would monitor, and what you would change with more time.

Say the step out loud as you move: "Let me do some quick estimates before I draw." It shows structure and keeps the interviewer with you.

Where a copilot helps, and where it does not

A system design copilot listens to the interviewer and, if it can see your board, follows what you have drawn. The useful ones keep a single design plan through the round and, for each question, suggest a short thing to say now, what to draw next, a key number to mention, and how to handle a challenge.

What it cannot do is understand the design for you. If you draw a cache because a suggestion said so, the first "why there?" will expose it. Use suggestions to remember a step or a number, not to make decisions you cannot defend.

Whiteboards and shared boards

Remote rounds use a shared board such as Excalidraw, Miro, FigJam, Google Drawings or the drawing pad in CoderPad. A copilot can only follow the board if it can see it, through a shared tab or window or a photo of your paper. Help both the interviewer and any tool by keeping the board tidy:

• Label every box with a plain name, such as "API service", not just an icon.
• Write the key numbers in a corner and update them.
• Number the arrows when you walk through a request.
• Leave room on the right for the deep dive.

Handling pushback and deep dives

Interviewers push back on purpose. "What happens if this database goes down?" is not a sign you failed; it is the deep dive starting. A simple pattern works:

1. Agree with what is true: "Right, that is a single point of failure."
2. Give the fix and its cost: "A replica with automatic failover, at the cost of some extra latency on writes."
3. Offer the alternative: "If we can accept slightly stale reads, we could also read from the replica."

Common deep dives include caching and when it goes stale, splitting a database across machines, queues for slow work, and what happens when two users change the same thing at once. Prepare a paragraph on each.

Check the rules

Employers' rules on AI help in technical rounds differ. Some allow it, some ban it and many say nothing. Ask the recruiter if unsure, and if it is banned, use a copilot only for practice.

How OphyAI handles system design

OphyAI's System Design tool is on the Premium plan ($59 a month): 30 credits a live session and 20 a practice round. Its Live Copilot listens to the interviewer and reads your whiteboard from a shared tab or window or a phone photo. It follows the same six steps above, and for each question gives a short "Say this now" answer, what to draw next, the key number to mention, and how to handle pushback. It works with Excalidraw, Miro, CoderPad, Google Drawings or FigJam if the board is visible in a shared tab or window.

A mock round is 35 to 60 minutes on a built-in whiteboard with a voice AI interviewer, up to 3 hints and a scorecard (a 0 to 100 score and 7 skill scores). Both modes run on the web, with Chrome or Edge for call audio; the desktop app runs the Live Copilot. The dedicated tool is not in the phone apps, though the phone Copilot has a System design session type that listens and helps you reason out loud. See System Design, the OphyAI profile and coding interview copilot.