By Hey Mocky · Reviewed
Why system design answers have to be spoken
In a system design round, the diagram is only part of the evidence. The rest is what you say while you draw: why a component exists, which assumption you are relying on, and what you gave up to get a result. If you stay silent while sketching, the interviewer has to guess at your thinking.
Two candidates can draw similar boxes and still leave very different impressions. One sketches quietly and explains everything at the end in a rush. The other states the constraints first, explains each choice as it goes, and finishes with a clear decision. The second candidate is far easier to follow and to evaluate.
Silent practice also hides problems that speaking exposes. Watch for these:
Going quiet while drawing, then trying to explain the whole design in the last five minutes
Naming a technology before stating the problem it solves
Listing trade-offs without choosing one
Assuming traffic or data volume without saying so
Running out of time before covering failure cases
What interviewers tend to listen for
Each stage of a design round produces something you can say out loud. The table shows phrasing that makes each stage easy to follow.
| Stage | What to say out loud | What it shows |
|---|---|---|
| Requirements | "Before I design, I'll confirm three functional needs and two quality targets." | You can scope an open-ended problem |
| Assumptions | "I'm assuming reads far outnumber writes, so I'll optimize the read path first." | You reason from stated inputs |
| Components | "The API validates input, the store holds the mapping, and a cache absorbs repeat reads." | Your structure is deliberate |
| Trade-offs | "A single database is simpler to operate. I'd split it only if writes outgrow one node." | You make decisions, not just lists |
| Summary | "The design meets the requirements. The main risk is X, and I'd revisit Y at larger scale." | You can close with a decision |
A five-step practice routine
Run this routine once per prompt. Keep a timer visible and speak every step, even when you are alone in the room.
Step 1: Pick one prompt and set a timer
Choose a single prompt, such as "design a URL shortener" or "design a notification system for order updates." Stay with it for the whole session, even if you think you already know the answer. Match the session to the length of the interview you are preparing for. If you don't know that yet, a 35-minute block is a reasonable starting point.
Write down only the prompt before you start. Save solution guides for after the session, so you practice producing an answer instead of recognizing one.
Step 2: State requirements and assumptions aloud
Spend the first few minutes asking questions out loud. Separate functional requirements (what the system must do) from non-functional ones (how well it must do it, such as latency, availability, or durability). Name each assumption you make about users, traffic, or data size, and flag the ones you would confirm with the interviewer.
Spoken example: "I'll assume users are spread across regions and that reads far outnumber writes. If either assumption is wrong, the caching plan changes."
Step 3: Sketch components and narrate one request
Draw the simplest version that meets your requirements. Then say what each component does and trace one request through the system. Describe each hop in a single sentence: "The client sends a create request to the API, which generates a short key, writes the mapping, and returns the link."
Step 4: Discuss trade-offs for each key decision
For every major choice, use the same four-part pattern: the choice, the alternative, the cost of your choice, and why you still picked it. Saying the cost is what separates a trade-off from a feature list. If you can't name a cost, you probably haven't examined the decision closely enough.
Step 5: Summarize in one minute
Stop, time yourself, and recap in about 60 seconds. Cover what the design meets, the main decision and its cost, the main risk, and what you would change at a larger scale. If the summary runs long, drop the component list and keep the decisions.
Worked example: a URL shortener
The template below turns the routine into a page you can copy into a notes app. Fill in the blanks for each session.
PROMPT: Design a URL shortener.
1. REQUIREMENTS (2-3 min)
Must: create a short link, redirect to the original URL
Quality: redirects feel instant; links don't expire unexpectedly
Assumptions I'll state: ____________________
2. COMPONENTS (5-7 min)
API: ____________________
Storage: ____________________
Cache: ____________________
One request traced: ____________________
3. TRADE-OFFS (5 min)
Decision: ______ Alternative: ______ Cost: ______ Why anyway: ______
4. SUMMARY (1 min)
Meets: ______ Main risk: ______ Changes at larger scale: ______Here is how one filled-in decision might sound. This is an illustrative sample, not a model answer:
"For the short key, I'd encode a counter into characters. Uniqueness is easy to guarantee that way. The cost is that the counter becomes a point of contention and makes link volume guessable. Random keys avoid both problems but need a collision check on write. For a first version I'd take the counter approach and allocate ranges to each writer if write load grows."
If you state traffic numbers, make them clearly illustrative. "Assume about a thousand new links a day for an internal tool" is a practice assumption, not a benchmark, and it should be stated as one.
Checking your explanation without a real call
The goal is to find gaps before the real interview. Don't rely on how the session felt. Use two checks.
First, ask a peer to listen. A colleague who sits in, keeps time, and marks every moment they lost the thread gives you more useful data than your own impression. Ask them to note any term that appeared without a reason, and any summary that repeated the diagram instead of the decisions.
Second, run each session through this checklist:
| Check | Done? |
|---|---|
| I stated at least two requirements before drawing | |
| I named every assumption about traffic or data size | |
| I explained why each component exists | |
| For each major decision, I named one alternative and its cost | |
| I covered at least one failure case | |
| My summary took about a minute and ended with a decision |
Then repeat the same prompt a few days later. If the same gap shows up twice, it is a habit to fix, not bad luck.
Where a voice mock interview fits
Once the routine feels familiar, the next step is to speak to something that responds. Hey Mocky is a practice portal for software engineers who want to rehearse before a real interview. It is a practice tool, not a live interview assistant, so it is not meant for use during a real call. Its purpose is to show you how your explanation holds up before the interview that counts.
Based on the current product details, a session works like this:
Setup: about two minutes. Add your resume and job description as PDF, DOCX, or TXT, or paste the text. Uploaded resume and job description files are not kept.
Call: Google Meet, Microsoft Teams, or Zoom-style layouts, with four interviewer voices (Ava, Ethan, Priya, and Arjun) and beginner, intermediate, or advanced difficulty. Chrome or Edge is recommended, with headphones.
Length: up to 30 open-ended questions per run, and up to 30 minutes per paid interview. The call warns you two minutes before the end.
Answers: speak them or type them. Every spoken word appears in a live transcript.
Report: an overall score, scores by skill, strengths and focus areas, feedback on every question, missed points with a model answer, communication scores, and a PDF you can save.
For a voice-based check on clarity and structure, this feedback is the part to use. The report shows where an answer was unclear or unstructured, which is the gap your peer and checklist would catch. Hey Mocky covers several roles, and its software engineer mock interview practice page is the place to start. It does not replace your own architecture study, so pair voice sessions with timed design sketches on your own prompts.
Your first interview is free: up to 10 minutes, a full scored report, and no card needed. After that, you pay per interview with no subscription. Paid credit packs provide five interviews of up to 30 minutes each; check the pricing page for current prices in your currency. Credits never expire, and a credit is used only when you press Start interview. Setup, the lobby, and reloading during an interview don't use a credit, but ending a call early still uses it. The base price is in USD; the site can display a local-currency price for the selected country. Taxes, discounts, and the final amount are confirmed at checkout. See the pricing page for current details.
For a practice interface that resembles an upcoming video call, see the Google Meet-style rehearsal and Zoom-style rehearsal pages. These are practice layouts, not integrations into real interviews.
Frequently asked questions
How long should each practice session run?
Run one complete session at the length of the interview you're preparing for, then spend about the same time reviewing it. Two or three full sessions a week on different prompts usually teaches more than one long cram. Return to your first prompt after a few days to see whether your narration has improved.
Do I need a partner to practice out loud?
No, but a partner hears what you can't hear yourself: the step you skipped and the term you never explained. If you practice alone, keep the timer visible, speak as if someone is listening, and use the checklist above. When no peer is free, a voice mock interview that gives feedback on each answer can cover some of the same ground.
Is talking through a system design the same as explaining code?
Not quite. A coding round tends to focus on working code and how you reason about its complexity. A system design round asks you to justify choices across several components, and it rarely has one correct architecture. The habits overlap, especially stating assumptions and summarizing, but system design leans harder on trade-offs, so practice step four until it feels natural.
Can I use Hey Mocky during a real interview?
No. Hey Mocky is a practice tool, not a live interview assistant, and it is built for rehearsal before the real interview. Use it to build the habit of explaining your design clearly, then take the real interview on your own. The FAQ explains how the sessions work.
What happens to my resume and job description files?
Uploaded resume and job description files are not kept. If you would rather not upload files at all, you can paste the text instead. For questions about how your data is handled beyond this, the privacy policy sets out the details.
Conclusion
System design practice works when you speak the whole design: the requirements, the components, the trade-offs, and the summary. Pick one prompt today, run the five steps aloud with a timer, and score yourself with the checklist. When you want a voice-based rehearsal with feedback on every answer, create your account and start with the free first interview, which needs no card.