Question Bank — 92 system-design questions

The index. Cue sheets live in questions/<LETTER>-<archetype>.md, one file per archetype, each question anchored as ### SD-nn.

Interviewer: do not read a whole archetype file. Slice it:

grep -n '^### SD-08' questions/B-social-feed-content.md   # → line

then Read with offset and limit ≈ 90. Reading the whole bank index is cheap; reading a whole archetype file is not.

Candidates: don’t read the cue sheets before a round — they contain the answer. This index is safe.

How the bank is organised — four ways to slice it

Axis Values Use it to…
Archetype (A–K) the technical family — what the round really tests get coverage; one round per archetype beats ten in your comfort zone
Kind Build (design X) · Review (inverted: critique an artifact) · Strategy (decide, sequence, invest) match the round format your loop uses; senior levels see more Review and Strategy
Domain general · fintech · crypto · ai · social · commerce · media · search · mobility · saas · infra · data rehearse in your target industry — and, deliberately, outside your home turf
Levels Sr senior · St staff · Pr principal · EM · SEM senior EM · Dir director pick questions that load your target level’s bar — see levels.md

Tiers. T1 = full cue sheet: crux, clarification numbers, checklist, phase-by-phase probe ladder, curveball, weak/strong/exceptional, traps. T2 = compact sheet: prompt, numbers, checklist, five probes, three bands — enough to run a complete round. 40 T1, 52 T2.

Asked by — where the question shows up: G Google · M Meta · A Amazon · MS Microsoft · N Netflix · U Uber/Lyft/DoorDash-class · FT fintech (payment companies, neobanks, issuers, lenders) · CX crypto exchanges & custodians · AI AI labs and AI-platform companies · ★ = a flagship for that company class, the one you should assume will be asked.

Every question can be run at every level — the Levels column says where it is most revealing. Running SD-08 at director is legitimate; the interviewer will steer to org and cost within fifteen minutes (levels.md §3).

A · Ingest, streaming & real-time analytics → questions/A-streaming-analytics.md

ID Question T Kind Domain Levels Asked by Loads
SD-01 Real-time audience metrics (approximate unique counts) 1 Build data media Sr St EM G M A streaming, approximation, lambda split
SD-02 Trending now / Top-K leaderboard 1 Build social media Sr St EM G★ M★ read/write asymmetry, sliding windows, heavy hitters
SD-03 Ad click aggregation & billing counters 1 Build data St EM G★ M★ A exactly-once accounting, late data, reconciliation
SD-04 Metrics & monitoring pipeline 2 Build infra data Sr St EM G M A MS time-series storage, cardinality, downsampling
SD-05 Log search & distributed tracing backend 2 Build infra St EM G A N inverted index at volume, sampling, retention tiers
SD-06 A/B experiment platform 2 Build data St EM SEM M★ G N assignment consistency, metric pipelines, guardrails
SD-07 Event-driven platform / durable log (Kafka-like) 2 Build infra data Sr St Pr G A MS partitioning, delivery semantics, consumer groups

B · Social, feed & content → questions/B-social-feed-content.md

ID Question T Kind Domain Levels Asked by Loads
SD-08 News feed / home timeline 1 Build social Sr St EM M★ G fan-out write vs read, the celebrity problem, ranking
SD-09 Chat / messaging (1:1 + groups) 1 Build social Sr St EM M★ MS★ connection state, ordering, delivery receipts, offline
SD-10 Live comments & reactions on a live stream 2 Build social media Sr St M★ pub/sub fan-out, backpressure, shedding
SD-11 Photo & video sharing (Instagram-like) 2 Build social media Sr EM M★ blob + metadata split, thumbnails, CDN
SD-12 Notification & push platform 1 Build social Sr St EM M A MS multi-channel fan-out, dedup, preferences, rate control
SD-13 Presence & online status 2 Build social Sr M MS heartbeat scale, fan-out cost, staleness budget
SD-14 Social graph store (friends, follows, FoF) 2 Build social St Pr M★ G graph sharding, adjacency caching, bidirectional writes

C · Storage, data & consistency → questions/C-storage-data-consistency.md

ID Question T Kind Domain Levels Asked by Loads
SD-15 Globally distributed KV store / multi-region consistency 1 Build infra St Pr G★ A MS consensus, quorums, CAP honesty, replication
SD-16 Object store data plane (S3-like) 2 Build infra St Pr A★ G erasure coding, durability math, large-object writes
SD-17 Object storage control plane 2 Build infra saas St EM A G metadata at scale, quotas, lifecycle, tenancy
SD-18 File sync (Drive / Dropbox) 1 Build saas Sr St EM G★ MS chunking, dedup, conflict resolution, sync protocol
SD-19 Collaborative document editing 1 Build saas St Pr G★ MS★ OT vs CRDT, presence, offline merge, undo
SD-20 Distributed cache 1 Build infra Sr St EM M★ G A consistent hashing, invalidation, stampedes, hot keys
SD-21 Data warehouse / batch analytics platform 2 Build data St EM SEM A G N columnar storage, partitioning, orchestration, cost
SD-22 Change data capture & replication pipeline 2 Build data St EM A MS N log-based CDC, schema evolution, ordering, backfill
SD-23 Multi-tenant SaaS platform isolation 2 Build saas St EM SEM MS★ A G noisy neighbours, quotas, cells, per-tenant data

D · Infrastructure & control planes → questions/D-infrastructure-control-planes.md

ID Question T Kind Domain Levels Asked by Loads
SD-24 Distributed lock / coordination service 2 Build infra St Pr G★ A consensus, fencing, lease expiry, safety vs liveness
SD-25 Rate limiter (distributed) 1 Build infra Sr St EM G★ M★ A MS algorithms, distributed counters, cell-local vs global
SD-26 Feature flags & config delivery 2 Build infra Sr St EM M★ N G push vs poll, staleness, kill switches, targeting
SD-27 Authorization / permission graph (Zanzibar-like) 2 Build infra saas St Pr G★ MS graph evaluation, caching consistency, tuple storage
SD-28 Workflow orchestrator / distributed job scheduler 1 Build infra St EM A★ G MS durable state machines, retries, exactly-once effects
SD-29 Device-fleet control plane (IoT) 2 Build infra St EM G A MS device identity, reconciliation loops, cohort rollout
SD-30 API gateway / edge platform 1 Build infra St EM SEM A★ MS G authn/z at edge, routing, quotas, deploys, blast radius
SD-31 Service discovery & load balancing 2 Build infra Sr St G A N health checking, propagation delay, LB algorithms
SD-32 CI/CD & build platform (developer productivity) 1 Build infra St EM SEM G★ M MS build graph, caching, queueing, tenancy, DORA metrics
SD-33 Cluster scheduler / capacity manager (Borg-like) 2 Build infra St Pr G★ MS bin packing, priority/preemption, quota, bad-node ejection
SD-34 Secrets management & key rotation 2 Build infra St EM MS A G root of trust, rotation without downtime, blast radius
SD-35 Progressive deployment & automatic rollback 2 Build infra St EM SEM G N A canary gates, halt conditions, blast radius, rollback safety

E · Search, geo & ML → questions/E-search-geo-ml.md

See also archetype I for AI & LLM systems — SD-42 is the serving prerequisite for SD-63/SD-64.

ID Question T Kind Domain Levels Asked by Loads
SD-36 Web crawler 1 Build search Sr St G★ M frontier, politeness, dedup, freshness, traps
SD-37 Search indexing & serving 2 Build search St Pr G★ M A inverted index, sharding by doc vs term, ranking tiers
SD-38 Typeahead / autocomplete 1 Build search Sr St G★ M A tries, precomputation, latency budget, personalisation
SD-39 Nearby / ride matching (geo) 1 Build mobility Sr St EM U★ G M geo indexing, matching under contention, state machines
SD-40 Maps routing & ETA 2 Build mobility St Pr G★ U graph preprocessing, contraction, live traffic ingestion
SD-41 Recommendation serving & feature store 1 Build ai commerce St EM M★ N★ A candidate generation → ranking, feature freshness, skew
SD-42 LLM inference serving / RAG platform 1 Build ai St Pr EM G M MS A GPU batching, KV cache, vector retrieval, cost per token
SD-43 Abuse / fraud detection in real time 2 Build fintech ai St EM M A★ rules + model serving, latency budget, feedback loops

F · Transactions, payments & commerce → questions/F-transactions-payments.md

SD-77–82 are the fintech expansion — card issuing, KYC/AML, reconciliation, FX, payouts, credit. Pair with archetype J for digital-asset roles.

ID Question T Kind Domain Levels Asked by Loads
SD-44 Payment processing & double-entry ledger 1 Build fintech Sr St EM A★ G MS idempotency, exactly-once effects, sagas, reconciliation
SD-45 Payment gateway integration (multi-PSP) 2 Build fintech commerce Sr St EM A MS adapter design, retries against third parties, webhooks
SD-46 Digital wallet & balance transfers 2 Build fintech Sr St A G atomic balance moves, hot accounts, holds, audit
SD-47 Ticket / seat booking under contention 1 Build commerce Sr St EM A★ G MS holds vs locks, oversell, fairness, expiry
SD-48 Inventory & order management at scale 1 Build commerce St EM A★ reservation vs availability, distributed inventory, sagas
SD-49 Delivery dispatch & order lifecycle 2 Build mobility commerce St EM U★ A long-lived state machines, assignment, real-time tracking
SD-50 Subscription billing & usage metering 2 Build saas fintech St EM A MS G metering accuracy, proration, idempotent invoicing
SD-77 Card authorisation & issuer processing 1 Build fintech Sr St EM FT★ A MS holds, stand-in, auth ↔ clearing matching
SD-78 KYC / AML onboarding & sanctions screening 2 Build fintech Sr EM FT★ MS durable workflow, fuzzy screening, continuous re-screening
SD-79 Reconciliation & settlement engine 1 Build fintech St EM SEM FT★ A MS multi-pass matching, expectations, exception workflow
SD-80 Cross-border payments & FX 2 Build fintech St EM FT★ A locked quotes, exposure, prefunding, multi-currency ledger
SD-81 Payouts & disbursements to bank rails 2 Build fintech commerce Sr St EM FT★ A U exactly-once outbound, ambiguous rail status, returns
SD-82 Real-time credit decisioning (BNPL) 2 Build fintech ai St EM FT★ A explainable decisions under budget, loan servicing

G · Media & delivery → questions/G-media-delivery.md

ID Question T Kind Domain Levels Asked by Loads
SD-51 Video upload, transcode & streaming (YouTube) 1 Build media St EM G★ N★ M pipeline fan-out, ABR, CDN, cost per hour watched
SD-52 Live streaming at low latency 2 Build media St Pr M★ N ingest fan-out, latency-vs-resilience, catch-up
SD-53 CDN / edge cache design 2 Build media infra St Pr G N A hierarchy, origin shielding, invalidation, hit ratio economics
SD-54 Partner data-feed / large-file delivery 1 Build data media St EM G A N generation vs delivery split, versioning, deadlines, authz

H · EM & senior-manager rounds → questions/H-em-specific-rounds.md

The archetype manager candidates skip, and the one an EM loop is most likely to contain in some form. These are not “design X” questions — they test the work an engineering manager actually does. Several invert the round: you present something and they respond. SD-89–92 are the senior-EM / director expansion — strategy conversations where org shape and investment are the design.

ID Question T Kind Domain Levels Asked by Loads
SD-55 Design review — critique and improve a given design 1 Review general St EM SEM G★ A MS judgment, prioritising findings, influence, tone
SD-56 Migration — monolith → services, or datacentre → cloud 1 Strategy general St EM SEM A★ MS★ G sequencing, dual-write/backfill, rollback, org staging
SD-57 Capacity, cost & FinOps review 1 Review general EM SEM Dir A★ N G unit economics, cost drivers, the levers, forecasting
SD-58 Reliability & SLO design for an existing system 1 Review general St EM SEM G★ A N SLI selection, error budgets, alerting, on-call load
SD-59 Multi-team decomposition & API contracts 1 Strategy general Pr EM SEM G★ MS A Conway’s law, seams, contract versioning, dependency risk
SD-60 Platform consolidation / tech-debt strategy 2 Strategy general Pr SEM Dir MS A G portfolio judgment, deprecation, migration incentives
SD-61 Incident → design fix 2 Review general St EM A★ G N RCA discipline, systemic vs local fixes, prevention
SD-62 Build vs buy 2 Strategy general EM SEM Dir MS A G TCO, differentiation, exit cost, team capability
SD-89 Org & architecture for a new product line 1 Strategy saas SEM Dir G MS A Conway’s law by design, reuse vs build, sequenced hiring
SD-90 Reliability program after a major outage 2 Strategy general SEM Dir A★ G N the failure class, SLOs/error budgets, no blanket freeze
SD-91 Annual planning — roadmap, headcount, platform 2 Strategy general EM SEM Dir G MS A capacity model, explicit trade-offs, platform business case
SD-92 Cloud, vendor & multi-region strategy 2 Strategy infra Pr SEM Dir A MS G N conflicting asks quantified, tiered redundancy, leverage

I · AI & LLM systems → questions/I-ai-llm-systems.md

New. Designing around a component that is expensive, non-deterministic and occasionally wrong. Related: SD-41, SD-42, SD-43 in E.

ID Question T Kind Domain Levels Asked by Loads
SD-63 LLM gateway & multi-model router 1 Build ai St EM SEM G MS A AI quotas & chargeback, routing/fallback, PII egress control
SD-64 AI agent platform — tool-using agents 1 Build ai St Pr EM AI★ G MS delegated least privilege, durable execution, injection containment
SD-65 LLM evaluation & observability platform 2 Build ai St EM AI★ G M no ground truth, calibrated judges, change gating
SD-66 Distributed training & GPU cluster scheduler 1 Build ai infra St Pr AI★ G M MS gang scheduling, failure math, checkpoints, goodput
SD-67 Vector search at billion scale 2 Build ai search Sr St G M A AI ANN recall vs memory, filtering, freshness
SD-68 Content moderation — ML plus human review 2 Build social ai St EM SEM M★ G MS triage by severity × confidence, reviewer capacity, retro policy
SD-69 ML platform — notebook to production 2 Build ai data St EM SEM M G A N train/serve skew, point-in-time features, lineage
SD-70 AI code-completion backend 2 Build ai Sr St AI G MS latency budget, cancellation, context assembly

J · Crypto & digital assets → questions/J-crypto-digital-assets.md

New. Custody of irreversible value against an adversary, with an external ledger that can reorganise. Asked at exchanges and at fintechs adding digital-asset products.

ID Question T Kind Domain Levels Asked by Loads
SD-71 Exchange matching engine & order book 1 Build crypto fintech St Pr CX★ FT sequencer + log as truth, single-threaded books, replay
SD-72 Custodial wallet — hot/warm/cold keys 1 Build crypto St EM CX★ FT risk-tiered custody, policy at the signer, insider threat
SD-73 Deposit detection & crediting across chains 2 Build crypto Sr St CX reorgs, confirmation state machine, idempotent credits
SD-74 Blockchain indexer & on-chain data API 2 Build crypto data Sr St CX reorg-aware CDC, backfill, hot addresses
SD-75 Margin & liquidation engine 2 Build crypto fintech St CX FT incremental risk, liquidation-price index, volatility peaks
SD-76 Fiat on-ramp & proof of reserves 2 Build crypto fintech St EM CX FT reversibility asymmetry, Merkle liabilities, attestation

K · Staff+ technical-leadership rounds → questions/K-staff-technical-leadership.md

New. The IC-track counterpart to H. Strategy, API review, deprecation, rewrite-vs-refactor, a performance investigation, and arbitration — judgment exercised without a reporting line. Two are inverted (SD-84, SD-87).

ID Question T Kind Domain Levels Asked by Loads
SD-83 Technical strategy — a 3-year platform direction 1 Strategy general St Pr SEM Dir G★ MS A diagnosis, few bets, what to stop, adoption without authority
SD-84 API design review 1 Review general fintech St Pr EM G★ MS A one-way doors, idempotency, money types, contracts
SD-85 Deprecating a widely used internal system 2 Strategy general St Pr SEM G MS A migration without mandate, zero-cost path, long tail
SD-86 Rewrite vs refactor 2 Strategy general St Pr EM G MS A N hidden requirements, strangler, evidence for the call
SD-87 Performance investigation — the p99 regression 1 Review general Sr St Pr G★ M A hypotheses from evidence, tail latency, mitigate vs fix
SD-88 Arbitrating two teams’ competing designs 2 Strategy general St Pr SEM G★ MS A agreed criteria, finding the crux, decidable disagreement

Browse by level

The questions that are most revealing at each level — the ones where the level’s bar is the thing being tested. Full ladder in levels.md.

Level Most revealing (★ = start here) Also strong
Sr SD-25★ SD-20★ SD-38 SD-02 SD-47 SD-87 SD-09 SD-44 SD-67 SD-73 SD-77
St SD-15★ SD-19★ SD-28 SD-66 SD-71 SD-84 SD-87 SD-24 SD-27 SD-33 SD-42 SD-64 SD-79 SD-86
Pr SD-83★ SD-88★ SD-27 SD-33 SD-59 SD-60 SD-14 SD-15 SD-66 SD-71 SD-85 SD-92
EM SD-55★ SD-58★ SD-08 SD-44 SD-57 SD-59 SD-12 SD-28 SD-32 SD-63 SD-77 SD-79 SD-91
SEM SD-56★ SD-59★ SD-60 SD-90 SD-92 SD-23 SD-32 SD-63 SD-68 SD-79 SD-85 SD-88
Dir SD-89★ SD-91★ SD-90 SD-92 SD-62 SD-60 SD-57 SD-83

Questions tagged per level: Sr 30 · St 81 · Pr 25 · EM 59 · SEM 24 · Dir 8.

Browse by domain

Run at least one round in your target domain before a domain-specific loop, and at least one outside it — fluent vocabulary in your own domain hides weak reasoning (profile.md raises the bar on your home domain).

Domain Questions Core test in this domain
general SD-55 SD-56 SD-57 SD-58 SD-59 SD-60 SD-61 SD-62 SD-83 SD-84 SD-85 SD-86 SD-87 SD-88 SD-90 SD-91 judgment, sequencing, review — domain-neutral rounds
fintech SD-43 SD-44 SD-45 SD-46 SD-50 SD-71 SD-75 SD-76 SD-77 SD-78 SD-79 SD-80 SD-81 SD-82 SD-84 correctness under retry; you are not the only ledger
crypto SD-71 SD-72 SD-73 SD-74 SD-75 SD-76 irreversible value, key custody, reorgs
ai SD-41 SD-42 SD-43 SD-63 SD-64 SD-65 SD-66 SD-67 SD-68 SD-69 SD-70 SD-82 expensive, non-deterministic components; evaluation without ground truth
social SD-02 SD-08 SD-09 SD-10 SD-11 SD-12 SD-13 SD-14 SD-68 fan-out, the celebrity problem, abuse at scale
commerce SD-41 SD-45 SD-47 SD-48 SD-49 SD-81 contention for scarce inventory, sagas
media SD-01 SD-02 SD-10 SD-11 SD-51 SD-52 SD-53 SD-54 moving bytes cheaply — egress dominates
search SD-36 SD-37 SD-38 SD-67 precomputation under a latency budget
mobility SD-39 SD-40 SD-49 geo indexing, matching under contention
saas SD-17 SD-18 SD-19 SD-23 SD-27 SD-50 SD-89 tenancy, isolation, collaboration
infra SD-04 SD-05 SD-07 SD-15 SD-16 SD-17 SD-20 SD-24 SD-25 SD-26 SD-27 SD-28 SD-29 SD-30 SD-31 SD-32 SD-33 SD-34 SD-35 SD-53 SD-66 SD-92 control plane vs data plane, blast radius
data SD-01 SD-03 SD-04 SD-06 SD-07 SD-21 SD-22 SD-54 SD-69 SD-74 event time, exactly-once accounting, lineage

Browse by kind

  • Review (6) — Inverted — the interviewer presents an artifact with planted defects and you review it. Rules: top of H-em-specific-rounds.md. SD-55 · SD-57 · SD-58 · SD-61 · SD-84 · SD-87
  • Strategy (12) — Decide, sequence, invest — the “design” is a plan, a recommendation or an org shape. SD-56 · SD-59 · SD-60 · SD-62 · SD-83 · SD-85 · SD-86 · SD-88 · SD-89 · SD-90 · SD-91 · SD-92
  • Build (74) — Classic “design X”. Every other question.

Selection guidance

If a real loop is close, run the company flagships first:

Company class Run these
Google L6 EM / Staff SD-15, SD-36, SD-02, SD-25, SD-19, SD-32, and two from H or K (SD-58, SD-59 for EM · SD-84, SD-87 for Staff)
Meta E6 EM / Staff SD-08, SD-09, SD-20, SD-41, SD-10, SD-06, SD-68
Amazon L6/L7 SD-48, SD-44, SD-47, SD-28, SD-30, SD-56, SD-57, SD-61, SD-90
Microsoft Principal / M2 SD-23, SD-19, SD-09, SD-34, SD-56, SD-59, SD-63
Netflix SD-51, SD-52, SD-41, SD-35, SD-57
Fintech (payments, neobank, issuing, lending) SD-44, SD-77, SD-79, SD-81, SD-45, SD-80, SD-78, plus SD-84
Crypto exchange / custodian SD-71, SD-72, SD-73, SD-44, SD-75, SD-76
AI lab / AI platform SD-42, SD-63, SD-64, SD-66, SD-65, SD-67
Director / Head of Engineering loops SD-89, SD-91, SD-90, SD-92, SD-62, SD-60

If there is no loop yet, go for coverage: one round in each archetype (A–K) beats ten rounds in the two that feel comfortable. Progress-Log.md tracks this.

The questions that most reliably expose a gap at each bar — where strong ICs stall and managers should not, or the reverse:

  • EM bar: SD-58 (SLOs and on-call), SD-57 (cost), SD-59 (splitting work across teams), SD-56 (sequencing a migration).
  • Staff bar: SD-87 (diagnosis from evidence), SD-84 (one-way-door review), SD-15 (consistency honesty), SD-88 (making a disagreement decidable).
  • Director bar: SD-89 (Conway’s law by design), SD-91 (capacity is the constraint), SD-92 (exposing conflicting asks).

Don’t practise what is already strong. After each round the debrief names one weak dimension; the next question should load that dimension hard:

Weak dimension Questions that load it hardest
Requirements & scoping SD-55, SD-12, SD-21, SD-83 — all deliberately over-broad
Estimation SD-03, SD-05, SD-51, SD-57, SD-66 — the numbers dominate the design
Technical depth SD-15, SD-19, SD-24, SD-16, SD-71, SD-87 — nowhere to hide
Trade-off reasoning SD-08, SD-18, SD-20, SD-62, SD-67, SD-86 — each has two defensible answers
Operations / cost SD-58, SD-57, SD-35, SD-30, SD-63, SD-92
Ambiguity & pushback any T1 at brutal difficulty; SD-55, SD-56, SD-88 especially
Leadership signal SD-59, SD-56, SD-60, SD-32, SD-85, SD-89

Adding a question

  1. Build the sheet with Cue-Sheet-Template.md.
  2. Append it to the right archetype file as ### SD-nn · Title [T1 · G★ M]. IDs are global and monotonic — take the next free number (currently SD-93) even if it lands in an older archetype; the archetype file’s header lists its ranges.
  3. Add a row here with Kind, Domain and Levels.
  4. Open a pull request — see CONTRIBUTING.md.

Table of contents