Module 20 / 20 · FINALE · Phase D — Build & Interview · 50 min

The interview
toolkit.

Everything from modules 1-19, distilled into a working playbook. A 6-phase framework, a question bank, the most common mistakes — and a simulator that walks you through any system design problem the way a senior engineer would. The last module. The one you'll come back to.

// What you'll know by the end

  • The 6-phase interview framework
  • Model answers for 5 canonical questions
  • The talk-track that separates good from great
  • The 4 most common mistakes (and the fix)
§ 01 — What the interview is really about

Not the
"right answer."

If you walk into a system design interview expecting to be tested on whether you know "the right way" to design Twitter, you've already lost. There is no single right way. The interviewer isn't comparing your answer to a key in their head. They're watching how you think when handed an ambiguous, large, real-world problem — exactly the kind you'll handle in the actual job. The interview is theater, but it's theater for a very specific skill: structured thinking out loud, under uncertainty.

// THE MYTHS · AND THE REALITY
WHAT CANDIDATES THINK
WHAT THE INTERVIEWER WANTS
i
"They want me to draw the perfect architecture."
They want to see you reason about tradeoffs. There is no perfect; only choices and consequences.
ii
"I should memorize the design for Twitter, Uber, WhatsApp."
They want to see you derive the design from the requirements. Memorization gets caught in 30 seconds.
iii
"More boxes = more impressive."
A simple correct system shows judgment. Over-engineering at the whiteboard tells them you'd over-engineer in production.
iv
"I should jump straight to the solution to show competence."
They want to see you scope first, then design. Jumping to architecture is the #1 red flag they screen for.
v
"I have to know everything they ask."
Saying "I don't know but here's how I'd find out" is a stronger signal than confidently guessing wrong.

That last one matters a lot. Strong candidates frequently say "I'm not sure" — about a specific technology, a specific number, a specific tradeoff. Then they describe how they'd resolve the uncertainty (read the docs, run a benchmark, ask the team). Weak candidates fake certainty because they think it's expected, and get caught when the interviewer probes a layer deeper. Calibrated confidence — knowing what you know and what you don't — is what's actually being measured.

The interview tests how you think under uncertainty, not what you've memorized.

So what does a structured thinker do at the whiteboard? They follow a framework — a sequence of phases that ensures they hit the right concerns in the right order. That framework is exactly what we'll spend the rest of this module on. Once you've internalized it, every "design X" question becomes the same exercise: walk the phases, fill in the details.

§ 02 — The 6-phase framework

Six phases.
Same every time.

Every well-conducted system design interview follows roughly the same arc. The phases below are the canonical decomposition — taught at FAANG companies, used in mock-interview books, and (most importantly) the actual structure senior engineers use in real production design reviews. The order matters. Do them out of order and the rest of the interview drifts.

// THE 6 PHASES · ROUGHLY HOW LONG EACH TAKES IN A 45-MIN INTERVIEW

1
Clarify requirements// scope the problem
Functional (what does it do?) and non-functional (how fast, how reliable, what scale?). Establish what's in scope and out of scope. Never skip this step.
~5 min
2
Back-of-envelope math// size the problem
DAU, QPS (peak), storage size, bandwidth. The numbers don't have to be precise — they have to be the right order of magnitude to drive architectural choices.
~3 min
3
API design// the interface
List the endpoints / RPC methods. Their inputs and outputs. POST /tweet, GET /feed. Forces you to commit to what the system actually exposes.
~5 min
4
Data model// what gets stored
The tables (or collections) and their key fields. Indexes. Relationships. Data shape constrains architecture — this informs everything in Phase 5.
~5 min
5
High-level architecture// draw the boxes
Web → API → cache → DB → queue → worker. The C4 Container view from M.19. Build up from naive to scaled, justifying each addition.
~15 min
6
Deep dive & tradeoffs// the senior-level questions
Pick one or two interesting pieces (the feed algorithm, the sharding scheme, the consistency model) and go deep. Discuss tradeoffs, not just the chosen path.
~10 min

A few things to internalize about this framework. First, the time estimates are loose — they shift based on the question and the interviewer's interests. But the relative weight is right: Phase 5 (architecture) is the biggest chunk, and Phase 6 (deep dive) is where senior signals land. Spend 20+ minutes on Phase 1, and you'll be rushed through the parts that actually matter.

Second, Phase 1 is non-negotiable. Even when you "know" the question is about Twitter, you still clarify: are we doing the feed, or the post step, or both? Public or private accounts? Just text or also media? The clarification questions are the answer to "do you scope problems before solving them" — which is what the interviewer is measuring.

Third, narrate the framework out loud. Tell the interviewer: "Let me start with requirements, then I'll do some back-of-envelope math, then sketch the API, data model, and architecture, and at the end I'll dive into whichever part you want to explore." This single sentence at the start of the interview signals you have a framework. It is shockingly effective.

§ 03 — Interview walkthrough · interactive simulator

Walk through
five real questions.

Below: pick one of five canonical system design questions. Step through the 6 phases. Each phase shows the prompts you should be hitting plus a model answer specific to that question. Switch questions to see how the same framework adapts. This is the actual playbook — bookmark it before your next interview.

INTERVIEW_WALK.SIM // m.20 finale
1REQS
2MATH
3API
4DATA
5ARCH
6DIVE
// THE FRAMEWORK PHASE 1
Clarify requirements
"What does this system do? At what scale?"
Loading...
// PROMPTS TO HIT
// MODEL ANSWER ·Design Twitter
§ 04 — The four most common mistakes

Where good
candidates slip.

The framework above gets you 80% of the way. The last 20% is avoiding the failure modes that even experienced engineers fall into when they're nervous. Each of these has a simple counter-move — and just knowing the trap exists halves your chance of falling into it.

// MISTAKE 1

Skipping requirements

"Cool, let me design Twitter — I'll start with the architecture..."
The fix: Always spend the first 5 minutes asking clarifying questions, even if you "know" the answer. Functional and non-functional requirements. In-scope and out-of-scope. "Just to confirm what we're building..." is a phrase that gets you points every time.
// MISTAKE 2

No tradeoff language

"We'll use Postgres. Done."
The fix: Present every choice as a tradeoff. "I'm choosing X over Y because [reason], though Y would be better for [other case]." This signals you considered alternatives. Senior engineers don't pick best; they pick best for these requirements.
// MISTAKE 3

Drawing the final state first

[Draws 12 services, 4 databases, 3 queues, and a CDN before saying anything]
The fix: Start naive. "Simplest version: one app server, one database." Then identify what breaks at scale and add components to fix it. Each addition justified by the previous bottleneck. This is exactly what M.17 taught.
// MISTAKE 4

Silent thinking

[Long pause, staring at the board, occasionally muttering]
The fix: Narrate everything. "I'm thinking about whether to shard by user_id or by tweet_id... If we shard by user_id, then the user's tweets stay together which is good for the feed query, but..." Even false starts are signal. Silent thinking shows nothing to evaluate.

One more anti-pattern worth flagging: defending your first answer at all costs. If the interviewer pushes back — "what about hot keys?", "what if traffic 10×s?", "how do you handle the celebrity case?" — they're inviting you to refine the design, not catching you in a lie. Welcome the pushback. Update the design out loud. "Good catch, you're right — in that case I'd add..." is one of the best things you can say. Stubbornness reads as lack of judgment; flexibility reads as senior.

§ 05 — How it all connects

Twenty modules.
One toolkit.

Before the final quiz, take a moment to see how all 20 modules link to the framework you just learned. Every concept from this track shows up somewhere in a real interview. The "deep dive" phase, especially, is where the deeper concepts pay off — that's the moment the interviewer is deciding whether to push for the next level up.

// THE BEGINNER TRACK · WHAT YOU NOW HAVE

// PHASE A · REQUEST & RESPONSE
The fundamentals every system uses

M.01 Systems M.02 Client-Server M.03 HTTP/HTTPS M.04 REST APIs

// PHASE B · DATA & STORAGE
Where data lives, why it's fast

M.05 SQL vs NoSQL M.06 Indexes M.07 Caching M.08 Latency

// PHASE C · SCALE & RELIABILITY
Going from prototype to production

M.09 Scaling M.10 LB M.11 DNS M.12 CDN M.13 Stateless M.14 Auth M.15 Obs M.16 Reliability

// PHASE D · BUILD & INTERVIEW
Designs, diagrams, and the toolkit

M.17 URL Shortener M.18 Image App M.19 C4 Diagrams M.20 Interview

Every "deep dive" topic in an interview maps to one of these modules. "How would you cache the feed?" — Module 7. "What if the DB falls over?" — Modules 9, 16. "How do you keep sessions across servers?" — Module 13. "Tell me about CDN caching policies" — Module 12. The list goes on. You don't have to recall all 20 modules during the interview; you just have to recognize which one applies to the question you're being asked.

§ 06 — Eight words for the interview room

Vocabulary,
for the whiteboard.

These show up in every system design conversation — interviews and real design reviews alike.

Functional vs Non-Functional
/ˈfʌŋkʃənl/
Functional = what the system does (endpoints, features). Non-functional = how well (latency, availability, scale). Both belong in Phase 1.
DAU / MAU
/diː eɪ juː/
"Daily Active Users" and "Monthly Active Users." The standard scale anchors. Phase 2 starts from one of these.
QPS
/kjuː piː ɛs/
"Queries per second." Always peak, never average. Drives sizing of caches, replicas, and queues.
Hot Key / Hot Shard
/hɒt kiː/
A single key or shard receiving disproportionate traffic (celebrity tweets, viral content). Always raised in Phase 6 deep dives.
Fan-out
/ˈfæn aʊt/
Distributing one event/write to many recipients. Push fan-out = pre-compute at write time. Pull fan-out = compute at read time. The classic feed-design tradeoff.
Eventual Consistency
/ɪˈvɛnʃʊəl/
A write becomes visible to all reads eventually, not immediately. Most large-scale systems accept it for non-critical data; flag where you do and don't.
Pre-aggregation
/priː ˌæɡrɪˈɡeɪʃən/
Computing expensive aggregations ahead of time (likes count, follower count) so reads stay fast. Often done with async workers updating denormalized counters.
Sharding
/ˈʃɑːdɪŋ/
Splitting a database across machines by some key (user_id, region). Unlocks horizontal scale for writes. The deep-dive topic for any "design X at billion-user scale."
§ 07 — The final knowledge check

Five questions.
One last quiz.

The last test of the track. Click an answer; explanation drops in instantly.

QUESTION 1 OF 5
Loading question...
Score: 0 / 5
5 / 5

Track complete.

The toolkit is yours. Twenty modules, one playbook, ready when the next whiteboard question comes.

§ 08 — The final recap

Three ideas
to carry forward.

If you take only three things from the whole track, take these.

i

Framework beats memorization

Six phases, same every time. Knowing them well beats memorizing the "right answer" for any specific question — because the right answer comes out of the framework.

ii

Tradeoffs are the answer

There is no best system. There's only best-for-these-requirements. Senior engineers narrate tradeoffs out loud; junior engineers state choices as facts.

iii

Calibrated confidence wins

"I don't know but here's how I'd find out" beats confident-but-wrong every time. The interview measures judgment, not omniscience.

◈ BEGINNER TRACK COMPLETE

You did it.

Twenty modules. Four phases. Dozens of interactive labs. From "what's a system?" to "design Twitter at scale" — the full arc. You now carry the vocabulary, the patterns, and the framework that opens every senior-level conversation in software engineering.

20
modules
20+
interactive labs
100
quiz questions
~18hr
of content

Now go write some better software.

↓ WHEN YOU'RE READY

The Intermediate track
awaits.

Distributed consensus. Event-driven architectures. Microservices boundaries. Stream processing. The patterns that hold real systems together. 26 modules. Same depth, same labs, same incremental approach — but at a level that gets you from "capable engineer" to "can be trusted to design new things."

Coming soon