Question to Offer
From a question
to an offer.
System design interviews at twenty-four companies. Real questions someone got asked. The hidden grading criteria the panel agrees on but never writes down. Worked deep dives, take-homes with rubrics, and the cross-company primitives that recur in every loop.
If any of this sounds familiar
You drew the right diagram.
They still passed on you.
- You walked out of a system design round and spent the train home replaying it, trying to find the thing you missed.
- You’ve been told you’re “almost” senior with no specific feedback on what that means.
- You’ve read every public system design resource and noticed they’re all the same six questions. None of them was the one you got asked.
- You had a take-home rejected with one line of feedback and no rubric. You still don’t know what the bar was.
- A coworker with worse technical chops got the L6 offer because they spoke fluently about tradeoffs you also knew.
None of that is bad luck. You got scored on something you couldn’t see. The diagram was fine. The conversation around it wasn’t. This course names what the panel was actually grading and shows you the answer that gets you the offer.
What you walk away with
The shape of any new question, the first time you see it.
Chapter 0
The five signals every panel scores you on
Ch 1–24
Twenty-four companies, real questions, real depth
Every unit
Ten-beat skeleton that mirrors a real onsite
End of every chapter
Fourteen take-homes with green/yellow/red rubrics
Ch 0.8
The hidden trade-off sentence senior engineers say
Chapter 25
The cross-company primitives that recur in every loop
Chapter 26
The panel meta-game: silence, hints, follow-ups, offer
Throughout
The vocabulary to defend a choice over the next-best option
Read enough of these and you start recognising the shape of any new question the first time you see it. That’s the actual outcome, and it’s the thing you’re paying for.
Twenty-four companies
One chapter each. Real questions. Real grading.
Each company chapter opens with how that company interviews — panel composition, level expectations, the cultural tells — then walks through six to eight specific questions they actually ask, plus a take-home.
The unit shape
Every interview unit, the same ten beats.
The ten beats mirror how a real onsite unfolds. You learn to recognise each one and what the interviewer is grading you on at each beat.
- 1. The question. The exact phrasing someone got it. Not paraphrased.
- 2. What’s actually being tested. The hidden signals behind the question.
- 3. Requirements (first 5 minutes). The cuts a strong candidate makes.
- 4. Back-of-envelope. Two minutes of numbers. Real arithmetic, shown.
- 5. High-level design. The diagram, drawn the way the panel reads it.
- 6. The deep dives that decide the grade. Two or three technical decisions, each defended over the next-best option.
- 7. Failure modes and 10x. What breaks at scale. What the senior interviewer asks at minute twenty-five.
- 8. The take-home version. The same question, asked as a multi-day deliverable.
- 9. Green / yellow / red rubric. What gets you the offer. What gets you a maybe. What gets you no.
- 10. Same question, different company. How the same answer would be graded differently at three other companies.
~20 minutes per onsite unit. Take-home units are longer (8–24 hours of practice work).
Syllabus
27 chapters · 221 units
Chapter 0 (foundation) is mostly open to read. Each company chapter starts with an intro on how that company interviews. 28 open units in total — enough to read every company’s playbook before deciding to unlock.
Ch 0FoundationFoundation11 units
Every panel scores the same five signals — scoping, estimation, design, depth, tradeoff. Vocabulary is leverage.
Ch 1GooglePEAK8 units
Platform-scale problems with a strong systems bias. Show the partition, show the failure mode.
- 1.0How Google interviews
- 1.1Web crawler at planet scale
- 1.2Google Docs real-time collaboration
- 1.3Google Drive multi-device sync
- 1.4Google Maps tile serving and routes
- 1.5Bigtable — row locality and tablet splits
- 1.6Distributed counter — ad impressions
- 1.7Take-home — key-value store that survives a node loss
Ch 2Meta / FacebookPEAK8 units
Social-graph scale and product velocity. The grade is in how you trade off fan-out and freshness.
Ch 3Amazon (retail)Medium8 units
Bar-raiser culture. Leadership Principles are scored on technical questions, not separately.
Ch 4MicrosoftMedium8 units
Enterprise scale and tenant isolation. The hard parts live in identity, sync, and permissions.
Ch 5AppleMedium8 units
Product-and-low-level hybrid. Privacy, on-device constraints, and silence as a signal.
Ch 6LinkedInMedium8 units
Graph + feed + jobs. Relevance is the recurring deep dive.
Ch 7YouTubePEAK8 units
Media pipelines at planet scale. Encoding, CDN, and counting are the deep-dive seams.
- 7.0How YouTube interviews
- 7.1Video upload, encoding, storage
- 7.2Recommendation system — the system side
- 7.3View counting — idempotent and anti-fraud
- 7.4Live streaming — HLS, DASH, and the latency-quality tradeoff
- 7.5Shorts — vertical short-form pipeline
- 7.6Comments at video scale
- 7.7Take-home — resumable multi-region upload service
Ch 8NetflixPEAK8 units
Reliability culture. Chaos and graceful degradation aren't extras — they're the rubric.
Ch 9TikTok / ByteDanceMedium8 units
Recommendation at cold-start scale. The hard parts are ingestion, ranking, and feedback loops.
Ch 10SnapMedium8 units
Ephemerality and geo are the design constraints, not features bolted on.
Ch 11SpotifyMedium8 units
Streaming and personalization. Offline sync and co-editing are the senior tells.
Ch 12RedditMedium8 units
Voting, ranking, and trees. The deep dive is anti-fraud and idempotent writes.
Ch 13X / TwitterMedium8 units
Fan-out is the question. Choosing write- vs read-time fan-out is the answer.
Ch 14UberPEAK8 units
Real-time marketplace. Dispatch, matching, and ETA are the recurring deep dives.
Ch 15AirbnbMedium8 units
Search, booking, and trust. Race conditions on availability decide the grade.
Ch 16DoorDashMedium8 units
Logistics in three-sided real time. Dispatch and payment splits are the seams.
Ch 17ShopifyMedium8 units
Commerce at burst scale. Black Friday is the implicit constraint on every answer.
Ch 18StripePEAK8 units
Correctness over cleverness. Idempotency, exactly-once-effect, and ledger integrity.
- 18.0How Stripe interviews — correctness first
- 18.1Idempotency keys
- 18.2Webhook delivery — retries, ordering, exactly-once-effect
- 18.3Card-on-file vault — PCI and tokenization
- 18.4Double-entry ledger
- 18.5Stripe Connect — multi-party transfers
- 18.6Radar — fraud detection at transaction time
- 18.7Take-home — rate-limited API gateway with idempotency
Ch 19CoinbaseMedium8 units
Money that can't be unspent. Custody, withdrawal safety, and price feeds are the dives.
Ch 20AWSPEAK8 units
Principled-engineer interviews. Show the failure model, show the consistency boundary.
Ch 21CloudflareMedium8 units
Edge-first. Cache propagation, isolates, and DDoS shape every answer.
- 21.0How Cloudflare interviews
- 21.1Global edge cache — purge propagation
- 21.2Workers — V8 isolates, near-zero cold start
- 21.3DDoS mitigation at L7
- 21.4Cloudflare Tunnel — private network through the edge
- 21.5R2 — object store with zero egress fees
- 21.6Durable Objects — stateful actors at the edge
- 21.7Take-home — edge cache with purge propagation
Ch 22DatadogMedium8 units
Observability at petabyte scale. Time-series, traces, and dedup are the recurring dives.
Ch 23FigmaMedium8 units
Multiplayer-first. CRDTs vs OT is the canonical deep dive.
Ch 24OpenAI / AnthropicMedium8 units
Serving and abuse at LLM scale. KV cache, batching, and rate limiting are the seams.
Ch 25Cross-company primitivesPEAK10 units
The patterns that recur in every loop. Name them once, recognize them everywhere.
- 25.0The primitives that recur
- 25.1Rate limiting — token bucket, sliding window, leaky bucket
- 25.2Consistent hashing
- 25.3Sagas and distributed transactions
- 25.4Distributed locking — and the Redlock controversy
- 25.5Leader election — Raft and Paxos
- 25.6Idempotency — the universal grading criterion
- 25.7CAP and PACELC in practice
- 25.8Bloom filters and probabilistic data structures
- 25.9Sharding strategies and rebalancing
Ch 26The meta-gameHigh point8 units
Panels, signals, the offer. The technical work is done; this is the rest of the grade.
The take-homes
Practice on the actual artifact.
Every company chapter ships a take-home. Some are real TypeScript code (a leader-elected job scheduler, a Reddit-style hot ranker, a rate-limited API gateway with idempotency). Some are six-to-ten-page design docs in the style the company actually writes them at — Amazon memo, Microsoft RFC, Apple security spec. You practice on the artifact, not on flashcards.
| Chapter | Shape | What you build / write |
|---|---|---|
| Code | KV store that survives a node loss | |
| Meta | Code | Write-heavy notification system |
| Amazon | Doc | Order tracking with state machine |
| Microsoft | Doc | Multi-tenant rate limiter (RFC) |
| Apple | Doc | iCloud account recovery |
| Code | Connection-strength scoring service | |
| YouTube | Doc | Resumable multi-region upload |
| Netflix | Code | Fault-injection middleware |
| Code | Hot-ranking service | |
| X | Code | Celebrity-aware fan-out |
| Shopify | Code | Shop Pay dedup service |
| Stripe | Code | Rate-limited API gateway with idempotency |
| AWS | Code | Leader-elected job scheduler |
| Cloudflare | Code | Edge cache with purge propagation |
| Figma | Code | Collaborative text editor (CRDT) |
Full access
Unlock the full course.
- ✓All 27 chapters, 221 units, every rubric and take-home
- ✓28open units — the foundation chapter and every company’s how-they-interview intro
- ✓Progress tracking and a continue-learning button, locally on your device
Frequent objections
Honest answers.
Who is this for?
Engineers with 3–10 years of experience preparing for system design interviews at top-tier companies. You should have shipped real production systems, have a working mental model of distributed systems vocabulary, and be staring down a real upcoming interview or two. You don't need to be senior to read it; you do need to be senior to make full use of it.
I've never interviewed at any of these companies. Will this help me?
Yes, with one caveat. The course teaches the shape of the interview, not company-specific trivia. If you're preparing for a smaller company that runs a similar loop, the foundation chapter, the cross-company primitives chapter, and the meta-game chapter will all transfer cleanly. Read the foundation chapter to check fit before unlocking.
What's a take-home unit?
Each company chapter ends with a practice take-home that mirrors what the company actually sends candidates. Some are real TypeScript scaffolds with tests (you build a hot-ranking service, a fault-injection middleware, a rate-limited API gateway). Some are design-doc prompts in the company's actual document style (Amazon memo, Microsoft RFC, Apple security spec). You practice on the artifact the panel asks for.
Why no specific model names in the AI labs chapter?
Same reason there are no SDK version numbers anywhere — model names age out within months. The Question to Offer course is written to last. The primitives that show up in an OpenAI or Anthropic interview today are the same primitives that will be on the next lab's interview a year from now. The course teaches those.
How is this different from leetcode practice?
It isn't, and it isn't trying to be. The coding round is real. We point at it once in the meta-game chapter and don't teach binary search. This course is the system design round, the take-home, the behavioral round technically, and the offer conversation. Pair it with a leetcode pattern book and you have the whole loop.
How long does this take?
Three to five hours to skim. Forty to sixty hours to read every unit. Hundreds of hours if you actually do the take-homes (which is the highest-ROI use of your time). Most people read the foundation chapter, the chapters for the companies they're targeting, plus the cross-company primitives and meta-game chapters. That's a focused twenty hours of reading.
Is the course updated when companies change their interview process?
The underlying patterns and grading signals are remarkably stable — most have been true for a decade. When a company makes a structural change (new round, new level, new ladder), we revisit the relevant chapter. You get the updates for life.
What if I read it and don't get the offer?
The course teaches what we can teach: the questions, the grading, the framework. We can't make you a better engineer in two weeks. If you read the foundation chapter and the chapters for your target companies and the round still goes badly, you'll have a clearer picture of what to study next — which is itself worth the time.
Why one price, no subscription?
Subscriptions feel exploitative for content that doesn't get consumed continuously. You buy it, you read it on your timeline, you keep it forever. The same applies if you come back to it in two years for a different interview round.
The foundation chapter is free.
Read the five-signals unit and the thirty-minute-clock unit. If they don’t change how you think about your next system design round, close the tab. If they do — the rest of the course is the catalog.