ClawPad
Sign inDownload
Menu
Architecture Question Bank & Blueprints

System Design Question Bank

Practice distributed-system reasoning with 100 categorized interview questions across 15 engineering domains. Explore worked blueprints, capacity calculations, and database schemas, then use a timed rehearsal to test your own explanation.

Start a 45-minute rehearsalBrowse the blueprints
100Total Questions
15Engineering Domains
20Deep Blueprints

One question · 45 minutes · One retry

Run a system-design rehearsal

Pick a question from the bank below. Keep its blueprint closed for the first 35 minutes. Use a timer and a blank sheet, alone or with a practice partner.

New to this format? Start with a URL shortener. Practicing a specific weakness? Choose a rate limiter for shared state or chat for message delivery. These are suggested drills, not predictions of an employer’s interview.

Choose a question from the blueprints
  1. 0–5 min

    Set the scope

    Name the users, two core operations, and one thing you will not build. State the latency and availability goals as assumptions to discuss.

  2. 5–10 min

    Estimate the load

    Choose a daily user count and requests per user. Convert to requests per second, explain a peak multiplier, and estimate storage with units.

  3. 10–22 min

    Draw and explain one path

    Sketch the simplest architecture on paper or a whiteboard. Trace one write and one read through the components. Name the stored records and their keys.

  4. 22–30 min

    Break one dependency

    Make a cache unavailable or delay a database replica. Trace what the user sees, how the system recovers, and what can be lost or duplicated.

  5. 30–35 min

    Defend one tradeoff

    Compare your choice with one plausible alternative. Explain which requirement wins and what new evidence would change your decision.

  6. 35–40 min

    Review the evidence

    Use the checklist below and then open a blueprint. Find one gap in your reasoning. Treat the reference as a design to question, not a script to memorize.

  7. 40–45 min

    Retry the weakest section

    Hide the reference and explain the corrected section aloud. Finish by choosing one changed constraint for your next session.

Review what you actually explained

For each row, mark missing, stated, or demonstrated. Demonstrated means you can point to your calculation, trace, or reason. Keep one note as evidence. This is a self-review rubric, not an automated score or a hiring prediction.

Five checks for your rehearsal. Scroll horizontally on a narrow screen.
CheckEvidence to look forFollow-up question
ScopeTwo core operations and a measurable goalCould another person tell what is out of scope?
CapacityArithmetic with units and stated assumptionsDo the numbers change a design choice?
Data flowA complete read and write path, including stored keysCan you explain where each piece of data lives?
Failure handlingA user-visible effect and a recovery pathWhat happens to a request already in flight?
TradeoffsAn alternative and a reason tied to a requirementWhen would the alternative become the better choice?

Example: if a cache outage makes every request hit the database, retry that failure explanation. State a load limit, decide which requests to reject or delay, and explain the user-visible result. Next time, repeat with ten times the traffic.

Use the capacity calculator to check your arithmetic after the first attempt. ClawPad can provide structured answer guidance and diagrams for study; you run the session and review your own reasoning.

Browse by Architectural Domain

🌐Social Media & Feeds7📡Streaming & Real-Time5💾Storage & Cloud Infrastructure6📍Geospatial & Mobility5🛍️E-Commerce & High-Concurrency5💳Financial & Payments5🔍Search & Big Data6⚡Distributed Infrastructure12🤖AI & Machine Learning5🚰Data Engineering & Pipelines5📊Data Science & Experimentation5🛡️Site Reliability & Platform5🔐Cloud Security & Zero Trust5🖥️Frontend & Client Systems5🤖GenAI & Autonomous Agents19
Complete Taxonomy

All 100 Questions by Category

Every classic and modern system design question asked across FAANG, fintech, high-growth startups, and AI labs.

🌐

Social Media & Feeds

(7 questions)

Fan-out architectures, feed ranking pipelines, follower graphs, and read-heavy caching.

Senior✓ Full Blueprint

Twitter / X Timeline & Fan-Out Engine

Design a high-throughput microblogging feed with fan-out-on-write vs fan-out-on-read for celebrity accounts.

Asked at
X / TwitterMetaLinkedInBluesky
Key Concepts
• Fan-out on write (push) vs Fan-out on read (pull)• Hybrid fan-out for celebrities (100k+ followers)• Redis timeline cache clusters• Snowflake 64-bit ID generation
Senior

Facebook News Feed & Ranking

Architect personalized social feeds incorporating social graph edge weights, affinity ranking, and infinite scrolling.

Asked at
MetaInstagramByteDancePinterest
Key Concepts
• Feed ranking algorithms• Graph databases• Feed generation pipeline• Distributed cache invalidation
Mid

Instagram Photo Sharing & Follower Graph

Handle massive image uploads with bidirectional follower graphs, asynchronous resizing, and real-time feed delivery.

Asked at
MetaSnapchatPinterest
Key Concepts
• S3 media storage• Distributed image resizing• Follower graph partitioning• Read-heavy caching
Senior

Reddit & Hacker News Threaded Discussion

Scale hierarchical comment trees, live upvote/downvote counters, and sub-community caching.

Asked at
RedditDiscordStack Overflow
Key Concepts
• Materialized path vs closure tables• Karma aggregation• Vote write coalescing• Subreddit sharding
Mid

Pastebin & Text Snippet Storage

Build an expiring text snippet storage system with collision-resistant unique short keys and automated TTL cleanup.

Asked at
AmazonGoogleGitHub
Key Concepts
• Base62 hash generation• Key generation service• S3 object storage• Automated TTL sweeps
Mid

Quora & Stack Overflow Q&A Platform

Design a high-volume question-and-answer platform with search tagging, view count deduplication, and reputation scoring.

Asked at
QuoraStack OverflowAtlassian
Key Concepts
• Full-text inverted index• View count deduplication via HyperLogLog• Tag hierarchy indexing• Reputation recalculation
Senior

Mint / Personal Finance Aggregator

Aggregate financial transactions from heterogeneous banking APIs with transaction categorization and audit logging.

Asked at
IntuitPlaidStripeRobinhood
Key Concepts
• Asynchronous webhook ingestion• Transaction categorization pipeline• Bank credentials vault• Idempotent deduping
📡

Streaming & Real-Time

(5 questions)

WebSockets, long-polling, adaptive video transcoding, audio streaming, and WebRTC.

Senior✓ Full Blueprint

Real-Time Distributed Chat (WhatsApp)

Deliver end-to-end encrypted instant messaging with bidirectional WebSockets, presence tracking, and offline message storage.

Asked at
MetaAppleTelegramSignal
Key Concepts
• WebSocket session gateway• User presence heartbeats• Cassandra/ScyllaDB message inbox• Push notification gateway
Staff✓ Full Blueprint

Slack & Discord Large Guild Messaging

Manage massive shared channels with live fanout to 100,000+ online members, message edit history, and reaction counters.

Asked at
DiscordSlackMicrosoft Teams
Key Concepts
• Guild member sync• Distributed pub/sub rings• Cassandra to ScyllaDB migration• Rate-limited reaction counters
Senior✓ Full Blueprint

Global Video Streaming Platform (YouTube / Netflix)

Ingest, transcode, and stream petabytes of on-demand video globally with adaptive bitrate streaming and multi-CDN distribution.

Asked at
NetflixYouTube / GoogleTwitch / AmazonTikTok
Key Concepts
• DAG transcoding pipelines• Adaptive bitrate (HLS / DASH)• Open Connect CDN edge caching• Video chunk manifest generation
Senior

Spotify Audio Streaming & Offline Cache

Stream compressed audio with instant playback start, playlist collaborative sync, and encrypted offline device storage.

Asked at
SpotifyApple MusicSoundCloud
Key Concepts
• Audio chunk prefetching• Peer-to-peer assisted CDN• Offline license validation• Real-time playlist CRDT sync
Staff✓ Full Blueprint

Zoom & Google Meet Video Conferencing

Low-latency multi-party video conferencing utilizing WebRTC, Selective Forwarding Units (SFUs), and bandwidth estimation.

Asked at
ZoomGoogleMicrosoftTwilio
Key Concepts
• WebRTC peer connections• Selective Forwarding Unit (SFU)• Simulcast & SVC video encoding• Audio mixer jitter buffers
💾

Storage & Cloud Infrastructure

(6 questions)

Distributed file systems, S3 blob stores, consistent hashing, and key-value stores.

Senior✓ Full Blueprint

Google Drive & Dropbox Cloud Storage

Sync files across desktop and mobile devices with block-level chunking, content-addressable storage, and conflict resolution.

Asked at
GoogleDropboxBoxMicrosoft
Key Concepts
• Block chunking & deduping (4MB chunks)• Content-addressed storage (SHA-256)• Delta synchronization protocol• Conflict forks
Principal

S3-Compatible Distributed Object Store

Store exabytes of unstructured blob data with 99.999999999% durability using erasure coding and distributed metadata clusters.

Asked at
AWSCloudflareGoogle CloudMinIO
Key Concepts
• Erasure coding (Reed-Solomon)• Data nodes vs metadata masters• Multipart upload coordination• Anti-entropy scrubbing
Staff✓ Full Blueprint

Distributed In-Memory Cache (Redis Cluster)

Provide sub-millisecond key-value lookups at millions of QPS using consistent hashing, virtual nodes, and active-passive failover.

Asked at
RedisAWS ElastiCacheMetaDatabricks
Key Concepts
• Consistent hashing ring• Virtual nodes distribution• Cache stampede thundering herd defense• LRU/LFU eviction algorithms
Staff

Distributed Key-Value Store (Dynamo / Cassandra)

Implement a highly available, partitioned key-value store with tunable consistency, vector clocks, and gossip-based failure detection.

Asked at
AmazonAppleUberNetflix
Key Concepts
• LSM-trees & SSTables• Vector clocks for conflicts• Sloppy quorum & hinted handoff• Gossip protocol membership
Staff

Content Delivery Network (CDN) & Edge Cache

Distribute web content with BGP Anycast routing, multi-tiered edge caches, and millisecond purge invalidations.

Asked at
CloudflareFastlyAkamaiAWS CloudFront
Key Concepts
• BGP Anycast routing• Origin shield caching• Instant cache purge propagation• HTTP Range requests for large files
Principal

Distributed File System (GFS / HDFS)

Manage petabytes of sequential analytical data across commodity hardware with single-master coordination and chunkserver replication.

Asked at
GoogleClouderaDatabricksSnowflake
Key Concepts
• Single active master architecture• 64MB chunk size rationale• Atomic record appends• Heartbeat & replica re-balancing
📍

Geospatial & Mobility

(5 questions)

Geohash & H3 indexing, quadtree partitioning, live driver dispatch, and routing algorithms.

Staff✓ Full Blueprint

Ride-Hailing & Real-Time Proximity (Uber / Lyft)

Match passengers with nearby drivers in real time with high-frequency GPS telemetry, dynamic surge pricing, and ETA routing.

Asked at
UberLyftGrabDoorDash
Key Concepts
• Uber H3 geospatial indexing• Driver location streaming buffer• Dispatch matching engine• Trip lifecycle state machine
Senior

Yelp / Google Places Proximity Search

Query millions of businesses by radius and bounding box with quadtrees, spatial indexing, and read-heavy caching.

Asked at
YelpGoogle MapsTripAdvisorFoursquare
Key Concepts
• Quadtree spatial partitioning• Geohashing vs PostGIS• Grid clustering at map zoom levels• High-read cache layers
Principal

Google Maps Turn-by-Turn Navigation & Routing

Compute shortest and fastest driving routes across global road networks with real-time traffic updates and vector tile rendering.

Asked at
GoogleAppleMapboxUber
Key Concepts
• Contraction hierarchies• A* search on road graphs• Real-time traffic overlay segments• Map vector tile hierarchies
Staff✓ Full Blueprint

DoorDash / UberEats 3-Sided Marketplace

Orchestrate real-time food delivery between customers, restaurants, and couriers with batched routing and live order tracking.

Asked at
DoorDashUberInstacartDelivery Hero
Key Concepts
• Batch dispatch optimization• Delivery ETA estimation• Order state machine orchestration• Real-time WebSocket tracking
Senior

Tinder Dating Match Engine & Swipe Queue

Process billions of daily swipes with geospatial distance filters, reciprocal match detection, and instant notification dispatch.

Asked at
Tinder / Match GroupBumbleHinge
Key Concepts
• Geospatial candidate pre-filtering• Bidirectional swipe queues• Bloom filters for seen profiles• Instant mutual match alerts
🛍️

E-Commerce & High-Concurrency

(5 questions)

Distributed inventory locks, flash sale throttling, double-booking defenses, and shopping carts.

Staff✓ Full Blueprint

Ticketmaster High-Concurrency Concert Ticketing

Handle viral concert ticket drops with virtual waiting rooms, seat reservation locks, and anti-scalping rate limits.

Asked at
TicketmasterSeatGeekEventbriteLive Nation
Key Concepts
• Virtual waiting room queueing• Distributed seat reservation locks (10-minute hold)• Optimistic concurrency control• Anti-bot CAPTCHA gates
Staff

Flash Sale & Limited Inventory Checkout

Sell 10,000 limited inventory units to 1,000,000 simultaneous shoppers without overselling or crashing downstream databases.

Asked at
AmazonShopifyAlibabaFlipkart
Key Concepts
• In-memory Redis atomic stock deduction• Asynchronous order queueing• Rate-limiting ingress gateways• Two-stage commit stock rollback
Senior

Airbnb & Hotel Booking System

Book lodging rooms across dynamic date ranges with overbooking prevention, calendar sync, and payment escrow holds.

Asked at
AirbnbBooking.comExpediaVrbo
Key Concepts
• Date range overlap queries• Pessimistic row locking on checkout• Idempotent payment capture• iCal distributed sync
Senior

Amazon E-Commerce Marketplace & Shopping Cart

Support multi-vendor item catalogs, persistent cross-device shopping carts, and inventory fulfillment workflows.

Asked at
AmazonShopifyWalmartTarget
Key Concepts
• Shopping cart session storage (DynamoDB)• Inventory soft reservation• Product search facet filtering• Fulfillment warehouse routing
Senior

Amazon Sales Rank by Category

Calculate and update sales ranks across millions of products in real time using rolling sliding window aggregations.

Asked at
AmazoneBayTarget
Key Concepts
• Sliding window counter aggregation• Logarithmic time-decay weighting• Category hierarchy rollup• Rank caching
💳

Financial & Payments

(5 questions)

Idempotent payment gateways, double-entry ledgers, low-latency matching engines, and fraud checks.

Staff✓ Full Blueprint

Idempotent Payment Processing Gateway (Stripe)

Authorize and settle multi-currency transactions with strict idempotency, third-party PSP failover, and ledger reconciliation.

Asked at
StripePayPalAdyenSquare / Block
Key Concepts
• Idempotency keys with Redis locks• Double-entry bookkeeping ledger• Asynchronous PSP reconciliation worker• Exponential backoff webhook dispatch
Staff

Digital Wallet & Double-Entry Accounting (PayPal)

Track user fiat and crypto balances with zero precision loss, ACID auditability, and Two-Phase Commit / Saga guarantees.

Asked at
PayPalVenmoCoinbaseRevolut
Key Concepts
• Double-entry ledger (Debits == Credits)• Saga orchestrator for multi-account transfers• Optimistic concurrency balance updates• Deterministic audit logs
Principal✓ Full Blueprint

Electronic Stock Exchange & Matching Engine

Trace FIFO priority, partial fills and cancellation in a hypothetical continuous limit book, then reason about sequencing, journal capacity and recovery.

Asked at
NasdaqCitadelJane StreetCoinbase
Key Concepts
• LMAX Disruptor ring buffer• In-memory price-time limit order book• Zero garbage collection techniques• UDP Multicast market data feeds
Senior

Retail Stock Brokerage (Robinhood)

Ingest live stock price feeds via WebSockets, place options and stock orders, and maintain real-time portfolio market value.

Asked at
RobinhoodInteractive BrokersCharles Schwab
Key Concepts
• WebSocket ticker streaming• Pre-trade risk & purchasing power check• Clearinghouse FIX protocol integration• Portfolio real-time mark-to-market
Staff

Real-Time Payment Fraud Detection Engine

Score payment card transactions in under 50ms using rule-based velocity limits, ML feature evaluation, and anomaly scoring.

Asked at
StripeVisaMastercardAmerican Express
Key Concepts
• Apache Flink real-time stream scoring• Sliding window card velocity checks• Low-latency feature store (Feast/Redis)• Dynamic blacklists/whitelists
🔍

Search & Big Data

(6 questions)

Web crawlers, prefix tries, inverted indexes, ad click stream aggregation, and heavy hitters.

Senior✓ Full Blueprint

Distributed Web Crawler

Crawl billions of web pages across the public internet respecting robots.txt politeness, duplicate URL detection, and DNS caching.

Asked at
GoogleBing / MicrosoftYahooBaidu
Key Concepts
• Crawl frontier priority queues• Host-based politeness delayers• Bloom filter duplicate URL elimination• Distributed DNS resolution cache
Senior

Google Search Autocomplete / Typeahead

Return top 5 search query suggestions in under 10ms as users type using distributed prefix tries and frequency ranking.

Asked at
GoogleAmazonBingDuckDuckGo
Key Concepts
• Trie data structure with top-k node caching• Offline frequency rollup via MapReduce/Spark• Prefix partitioning across servers• Browser local storage caching
Principal

Web Search Inverted Indexing & Ranking

Build an inverted index across billions of crawled HTML documents and serve ranked search results using BM25 and PageRank.

Asked at
GoogleMicrosoftElasticAlgolia
Key Concepts
• Inverted index posting lists• Document sharding vs term sharding• TF-IDF / BM25 relevance scoring• Tiered search index caching
Senior

Real-Time Ad Click Event Aggregator

Ingest and aggregate millions of ad click events per second for advertisers with exactly-once stream processing and minute rollups.

Asked at
Google AdsMeta AdsAmazon AdsThe Trade Desk
Key Concepts
• Kafka topic partitioning• Apache Flink tumbling and sliding windows• Watermark handling for late-arriving events• OLAP database storage (ClickHouse / Pinot)
Staff

Top-K Frequent Elements (Heavy Hitters)

Identify the top 100 most viewed videos or trending hashtags in real time using probabilistic streaming algorithms.

Asked at
Twitter / XYouTubeTikTokReddit
Key Concepts
• Count-Min Sketch• Space-Saving algorithm• Lossy counting• Min-heap maintenance with Redis ZSET
Mid

Real-Time Global Gaming Leaderboard

Maintain live player rankings for 25 million active gamers with score updates and instant percentile lookups.

Asked at
Riot GamesEpic GamesRobloxElectronic Arts
Key Concepts
• Redis Sorted Sets (ZADD, ZRANGE)• Score range sharding across Redis nodes• Tie-breaker timestamp encoding• Read replica caching for top 100
⚡

Distributed Infrastructure

(12 questions)

Rate limiters, distributed cron schedulers, snowflake IDs, message brokers, and APM tracing.

Senior✓ Full Blueprint

Distributed Rate Limiter & Token Bucket

Protect multi-region microservices from traffic spikes, brute-force attacks, and noisy neighbors with sub-millisecond overhead.

Asked at
CloudflareStripeAWSGoogle
Key Concepts
• Token bucket & sliding window log algorithms• Redis Lua atomic script execution• Local in-memory token buffering• HTTP 429 Too Many Requests response headers
Mid✓ Full Blueprint

Scalable URL Shortener (TinyURL)

Compress long URLs into 7-character aliases with high read-to-write ratios, fast HTTP 301/302 redirects, and analytics tracking.

Asked at
Bit.lyGoogleAmazonTwitter
Key Concepts
• Base62 encoding• Pre-generated key distribution service (KGS)• High-performance redirect caching• Asynchronous click analytics pipeline
Mid✓ Full Blueprint

Distributed Notification Engine

Fan out millions of transactional and promotional notifications across APNs, FCM, SMS, and Email with priority queues.

Asked at
TwilioUberAmazon SNSDoorDash
Key Concepts
• Multi-provider routing (APNs, FCM, Twilio, Sendgrid)• Priority queue partitioning• User notification preference matrix• Rate-limited deduplication
Senior✓ Full Blueprint

Real-Time Metrics Monitoring & Alerting (Datadog)

Collect, aggregate, and query millions of infrastructure metrics per second with configurable alerting thresholds and dashboards.

Asked at
DatadogPrometheusNew RelicGrafana Labs
Key Concepts
• Time-series database (TSDB) chunking• Push vs pull metric collectors• Streaming alerting evaluation engine• Data downsampling rollups
Staff✓ Full Blueprint

Distributed Job Scheduler & Orchestrator

Execute millions of recurring cron and delayed background jobs reliably with worker heartbeats and distributed locking.

Asked at
TemporalAirbnb / Apache AirflowAWS BatchMeta
Key Concepts
• Delay queues via Redis ZSET or Kafka• Worker heartbeat and lease renewal• At-least-once job execution• Dead-letter queues for failed executions
Mid

Distributed Unique ID Generator (Snowflake)

Generate 64-bit globally unique, roughly time-sorted integers across distributed worker nodes without database coordination.

Asked at
Twitter / XInstagramDiscordFigma
Key Concepts
• Twitter Snowflake layout (Epoch + Machine ID + Sequence)• NTP clock drift mitigation• Worker node ID assignment via ZooKeeper• Sequence overflow handling
Staff✓ Full Blueprint

Distributed Message Queue (Apache Kafka)

Follow a conceptual Kafka example through partitions, committed offsets and a consumer crash, then size a hypothetical distributed log.

Asked at
Confluent / Apache KafkaAWS SQS / KinesisRabbitMQApache Pulsar
Key Concepts
• Partitioned append-only commit logs• Zero-copy OS disk-to-network transfer (sendfile)• Consumer group rebalancing• In-sync replicas (ISR) and leader election
Staff

Distributed Lock Manager (Redlock / Chubby)

Synchronize access to shared resources across untrusted distributed processes with lease timers and fencing tokens.

Asked at
Google ChubbyRedisApache ZooKeeperHashiCorp Consul
Key Concepts
• Redis Redlock algorithm consensus• Fencing tokens to prevent split-brain zombies• Heartbeat lease renewals• Clock drift safety assumptions
Staff

Distributed Tracing & APM (OpenTelemetry / Jaeger)

Track end-to-end request journeys across hundreds of microservices with trace context propagation and adaptive sampling.

Asked at
DatadogSplunkDynatraceGoogle Cloud Trace
Key Concepts
• W3C Trace Context (traceparent header)• Tail-based vs head-based trace sampling• High-throughput collector buffer• ClickHouse span storage
Senior

LeetCode Online Code Execution Judge

Compile and run untrusted user code safely across 30+ languages with strict CPU/memory timeouts and vulnerability sandboxing.

Asked at
LeetCodeHackerRankCodeSignalGitHub Actions
Key Concepts
• Linux cgroups and namespaces isolation• gVisor / Firejail kernel security barriers• Asynchronous job submission queue• Memory and wall-clock enforcement
Senior

Webhook Delivery & Event Notification Engine (Svix)

Deliver billions of outbound HTTP webhooks to third-party customer endpoints with exponential backoff, signatures, and circuit breakers.

Asked at
SvixStripeShopifyGitHub
Key Concepts
• HMAC SHA-256 webhook signatures• Exponential backoff retry with jitter• Per-endpoint circuit breaking• Dead-letter queues and replay portals
Staff

Real-Time Collaborative Document Canvas (Figma / Docs)

Enable concurrent multi-user editing on documents and canvas trees with conflict resolution, presence cursors, and undo/redo.

Asked at
FigmaGoogle DocsNotionMiro
Key Concepts
• Operational Transformation (OT) vs CRDTs• WebSocket state sync• Presence cursor broadcast with throttling• Undo/redo action tree stacks
🤖

AI & Machine Learning

(5 questions)

LLM serving with continuous batching, vector search, recommendation feeds, and streaming ASR.

Staff

Video Recommendation Engine (TikTok / YouTube)

Serve personalized video feeds in under 50ms combining two-stage candidate generation, deep ranking models, and real-time user feedback.

Asked at
ByteDance / TikTokYouTubeMeta / Instagram ReelsNetflix
Key Concepts
• Two-stage candidate retrieval (ANN) and ranking• Real-time user engagement feature store• Exploration vs exploitation (Multi-armed bandits)• Cold-start handling
Staff

Visual Search & Image Embedding Engine (Google / Pinterest)

Search a catalog of 10 billion products by uploaded image using Vision Transformers, vector indexing, and real-time metadata filtering.

Asked at
Google LensPinterestAmazon Visual SearchAlibaba
Key Concepts
• Vision Transformer (ViT) embedding generation• Vector similarity search (HNSW / ScaNN)• Hybrid vector + metadata filtering• Embedding model versioning
Principal✓ Full Blueprint

High-Throughput LLM Inference Serving System (vLLM)

Serve generative Large Language Models with continuous batching, PagedAttention KV cache memory management, and speculative decoding.

Asked at
OpenAIAnthropicTogether AIGroqGoogle DeepMind
Key Concepts
• Continuous iteration-level batching• PagedAttention KV cache virtual memory paging• Tensor parallelism across GPUs• Speculative decoding verification
Senior

Enterprise RAG (Retrieval-Augmented Generation) Pipeline

Ingest enterprise documents, chunk and embed text into vector databases, and perform hybrid lexical-semantic retrieval for LLMs.

Asked at
OpenAICoherePineconeDatabricksAnthropic
Key Concepts
• Semantic document chunking• Dense vector embeddings + Sparse BM25 hybrid search• Cross-encoder re-ranking• Context window token budgeting
Senior

Real-Time Streaming Speech-to-Text Pipeline (ASR)

Transcribe live microphone audio in under 200ms latency using chunked streaming ASR, connection pooling, and speaker diarization.

Asked at
xAI / Grok-STTSonioxOpenAI WhisperDeepgramGoogle Speech
Key Concepts
• Opus audio chunk streaming• Streaming Conformer / Whisper inference• VAD (Voice Activity Detection) endpointing• Backpressure audio frame buffering
🚰

Data Engineering & Pipelines

(5 questions)

Lakehouse storage, real-time stream ingestion, Change Data Capture (CDC), and partition compaction.

Staff

Real-Time Clickstream Ingestion & Sessionization (Kafka + Flink)

Ingest 1M events/sec from mobile and web, deduplicate, sessionize with tumbling/session windows, and sink into analytical lakehouses.

Asked at
NetflixPinterestStripeSnowflake
Key Concepts
• Kafka topic partitioning• Apache Flink stateful stream processing• Watermarking & late-arriving event handling• Exactly-once semantics (2PC sinks)
Senior

Change Data Capture (CDC) Lakehouse Ingestion Engine (Debezium + Iceberg)

Stream transactional database mutation logs (PostgreSQL WAL) into Apache Iceberg tables with partition compaction and ACID guarantees.

Asked at
AirbnbUberDatabricksShopify
Key Concepts
• PostgreSQL WAL / MySQL Binlog streaming• Apache Iceberg metadata trees• Copy-on-Write vs Merge-on-Read• Automated small-file compaction
Senior

GDPR / CCPA Right-to-be-Forgotten Data Lake Purger

Erase specific user identifier records across petabytes of immutable historical Parquet files in S3 while minimizing write amplification and compute cost.

Asked at
MetaGoogleSpotifyAmazon
Key Concepts
• Bloom filters for record locating• Selective Parquet row-group rewriting• Secondary index lookup maps• Immutable audit trail generation
Staff

Financial Audit & Double-Entry Analytical Warehouse (Stripe / Square)

Reconcile billions of daily financial transactions across multiple payment rails with point-in-time snapshot auditability and zero ledger drift.

Asked at
StripeBlock / SquareRobinhoodAdyen
Key Concepts
• Immutable append-only ledger tables• As-of point-in-time time-travel queries• Multi-currency rounding & balance invariants• Automated cross-bank settlement reconciliation
Senior

High-Throughput Metric Rollup & Anomaly Pipeline (Datadog / Databricks)

Ingest 10M time-series telemetry data points per second, compute multi-tier downsampling rollups (1m, 1h, 1d), and detect statistical anomalies.

Asked at
DatadogDynatraceGrafana LabsAmazon CloudWatch
Key Concepts
• Time-bucketed aggregation windows• Gorilla XOR timestamp compression• Streaming exponential smoothing (EWMA)• ClickHouse materialized views
📊

Data Science & Experimentation

(5 questions)

Enterprise A/B testing platforms, switchback experiments, variance reduction (CUPED), and metric evaluation.

Staff

Enterprise A/B Testing & Metric Computation Platform (Statsig / Eppo)

Deterministically assign millions of daily active users into concurrent experiment buckets, evaluate metric shifts, and prevent sample ratio mismatches.

Asked at
StatsigEppoNetflixMetaBooking.com
Key Concepts
• MurmurHash deterministic user hashing• Layered experiment isolation grids• Sample Ratio Mismatch (SRM) chi-square test• Automated guardrail metric alerts
Senior

Marketplace Switchback & Cluster Experimentation Engine (DoorDash / Uber)

Measure pricing and dispatch algorithm changes in two-sided marketplaces without cannibalization or spatial network spillover.

Asked at
DoorDashUberLyftInstacart
Key Concepts
• Time-space switchback randomization• H3 geospatial cluster partitioning• Network interference mitigation• Bootstrap standard error estimation
Senior

Variance Reduction & Fast-Significance Engine (CUPED)

Accelerate A/B test runtimes by 50% using Controlled-experiment Using Pre-Experiment Data (CUPED) to remove pre-existing user variance.

Asked at
MicrosoftAirbnbNetflixAmazon
Key Concepts
• ANCOVA / Covariate adjustment• Pre-experiment feature extraction• Linear regression coefficient estimation• Statistical power calculation
Staff

Search Relevance & Ranking Offline Evaluation Harness (Google / Amazon)

Evaluate search ranking algorithms against human evaluation judgments and implicit click logs using NDCG, MAP, and MRR metrics.

Asked at
GoogleAmazonPinterestEtsy
Key Concepts
• Normalized Discounted Cumulative Gain (NDCG@K)• Mean Reciprocal Rank (MRR)• Interleaving search results for fast online eval• Position bias click models
Mid

Customer Churn Prediction & Early-Alert Pipeline

Ingest daily user product engagement signals, compute RFM (Recency, Frequency, Monetary) metrics, and score probability of 30-day account abandonment.

Asked at
SpotifySalesforceHubSpotLinkedIn
Key Concepts
• RFM feature engineering• Survival analysis (Cox Proportional Hazards)• Calibration curves & Brier score• Automated webhook trigger dispatch
🛡️

Site Reliability & Platform

(5 questions)

Multi-region active-active disaster recovery, distributed APM tracing, canary rollouts, and chaos injection.

Principal

Global Multi-Region Active-Active Disaster Recovery Architecture

Route traffic dynamically across three global cloud regions with zero-downtime automated failover, data synchronization, and split-brain immunity.

Asked at
AWSGoogle CloudNetflixCloudflare
Key Concepts
• BGP Anycast & Route 53 latency routing• CockroachDB / Spanner multi-region replication• Automated circuit breaker traffic shifting• RTO and RPO SLA guarantees
Staff

High-Volume Distributed Tracing & APM Platform (OpenTelemetry / Jaeger)

Ingest 1M spans per second across microservices, trace distributed context propagation, and execute tail-based anomaly sampling.

Asked at
DatadogDynatraceHoneycombLightstep / ServiceNow
Key Concepts
• W3C TraceContext header propagation• Head vs Tail-based sampling filters• DAG span assembly & dependency graph• ClickHouse columnar trace storage
Senior

Automated Canary Deployment & Rollback Engine (Argo Rollouts / Spinnaker)

Gradually shift production traffic to new service releases (5% -> 25% -> 100%), monitor live Prometheus error budgets, and trigger instant rollbacks.

Asked at
NetflixMetaIntuit / ArgoStripe
Key Concepts
• Envoy service mesh weighted routing• Prometheus Mann-Whitney U metric comparison• Automated GitOps rollback PR trigger• Database schema backward-compatible expand-contract
Staff

Multi-Cluster Kubernetes Secret & Config Synchronizer

Securely replicate dynamic credentials and configuration across 500 Kubernetes clusters with automated rotation and zero plain-text storage.

Asked at
HashiCorpShopifyDatadogTarget
Key Concepts
• Kubernetes Custom Resource Definitions (CRDs)• Vault dynamic secret engines• mTLS SPIRE cluster attestation• Envelope encryption with AWS KMS
Staff

Chaos Engineering Automated Fault-Injection Platform (Chaos Monkey / Gremlin)

Proactively inject simulated latency, packet drops, CPU exhaustion, and AZ outages into production clusters with automated blast radius containment.

Asked at
NetflixAmazonGremlinMicrosoft
Key Concepts
• Blast radius containment controls• eBPF kernel packet delay injection• Live steady-state metric hypothesis checks• Automated emergency kill switch
🔐

Cloud Security & Zero Trust

(5 questions)

Enterprise secrets management, zero-trust remote proxies, SIEM threat correlation, and E2EE protocols.

Principal

Enterprise Secrets Management & Key Broker Service (HashiCorp Vault)

Store and broker application API keys, database credentials, and cryptographic certificates with leasing, revocation, and envelope encryption.

Asked at
HashiCorpStripePalantirApple
Key Concepts
• Shamir's Secret Sharing unsealing• Envelope encryption (KEK + DEK)• Dynamic short-lived database credentials• Cryptographic audit logging
Staff

Zero Trust Remote Access Gateway (Cloudflare Access / BeyondCorp)

Replace corporate VPNs by authenticating employee identity, verifying MDM device posture, and proxying internal enterprise applications.

Asked at
CloudflareGoogle (BeyondCorp)ZscalerPalo Alto Networks
Key Concepts
• Identity-aware reverse proxying• Mutual TLS (mTLS) client certificates• Real-time device posture verification• Ephemeral signed JWT user headers
Senior

Real-Time SIEM Threat Detection & Event Correlator (Splunk / Panther)

Ingest 200k audit events/sec from cloud trail logs, firewalls, and endpoint agents to detect lateral movement and credential exfiltration.

Asked at
CrowdStrikeSplunkPanther LabsPalo Alto Networks
Key Concepts
• Log schema normalization (OCSF format)• Streaming detection rules (Python / Sigma)• Sliding-window entity risk scoring• Automated SOAR playbook trigger
Staff

End-to-End Encrypted (E2EE) Messaging Protocol (Signal Protocol)

Guarantee forward secrecy and post-compromise security for asynchronous one-to-one and group messaging across distributed devices.

Asked at
SignalWhatsApp / MetaTelegramApple (iMessage)
Key Concepts
• X3DH prekey exchange protocol• Double Ratchet symmetric key derivation• Forward secrecy & future secrecy guarantees• Encrypted out-of-band attachment storage
Principal

High-Performance IAM Policy Evaluation Engine (AWS IAM / OPA)

Evaluate complex attribute-based access control (ABAC) and role-based policies in under 1 millisecond for every internal microservice request.

Asked at
AWSGoogle CloudStyra (OPA)Okta
Key Concepts
• Explicit Deny > Explicit Allow precedence• Attribute-based access control (ABAC)• Pre-compiled WebAssembly policy evaluation• Local in-memory policy cache synchronization
🖥️

Frontend & Client Systems

(5 questions)

Real-time collaborative canvases (CRDTs), offline-first sync engines, media prefetching, and spreadsheets.

Staff

Real-Time Collaborative Whiteboard & Canvas (Figma / Miro)

Synchronize infinite vector canvas nodes, shapes, and presence cursors across dozens of concurrent editors with smooth interactive responsiveness.

Asked at
FigmaMiroCanvaAtlassian
Key Concepts
• Conflict-Free Replicated Data Types (CRDTs)• WebAssembly scene graph rendering (WebGL)• Presence cursor throttling & interpolation• Local undo/redo action tree management
Senior

Offline-First Mobile & Web Sync Engine (Linear / Notion)

Enable instant local writes on client devices without network connectivity and reliably resolve state conflicts upon reconnect.

Asked at
LinearNotionSuperhumanApple Notes
Key Concepts
• Local SQLite / IndexedDB persistence• Client mutation outbox queue• Last-Write-Wins (LWW) with Lamport timestamps• Delta state synchronizer over WebSocket
Senior

Infinite Scroll Media Feed with Dual-Player Prefetching (TikTok / Reels)

Deliver zero-latency vertical video swiping on constrained mobile devices with predictive chunk prefetching and strict memory caps.

Asked at
TikTok / ByteDanceInstagram / MetaYouTube ShortsSnapchat
Key Concepts
• Dual-player pooling & reuse• Predictive 2-second chunk prebuffering• LRU disk cache eviction with quota enforcement• Adaptive bitrate (ABR) bandwidth probing
Staff

High-Performance Virtualized Spreadsheet Grid (Google Sheets / Airtable)

Render 1,000,000 tabular rows at 60fps with cell formula dependency graphs, virtualized DOM scrolling, and Web Worker computation.

Asked at
GoogleAirtableMicrosoft (Excel Online)Smartsheet
Key Concepts
• DOM node virtualization (windowing)• Directed Acyclic Graph (DAG) formula evaluation• Offscreen Web Worker calculation thread• Dirty-rectangle canvas rendering
Senior

Enterprise Micro-Frontend Shell & Module Federation Platform

Orchestrate 30+ independently deployed frontend applications into a unified single-page shell with shared dependencies and sandboxing.

Asked at
AmazonSpotifyPayPalUber
Key Concepts
• Dynamic runtime script injection• Shared singleton vendor dependencies (React)• CSS isolation (Shadow DOM / Scoped CSS)• Cross-app event bus communication
🤖

GenAI & Autonomous Agents

(19 questions)

Agentic tool loops, real-time AI search, inference serving, semantic caching, and LLM evaluation.

Staff

AI Coding Assistant & Autocomplete Engine (Cursor / GitHub Copilot)

Provide inline code completions and repository-aware chat by indexing AST symbols, open editor tabs, and local git diffs.

Asked at
Cursor / AnysphereGitHubMicrosoftSupermavenAugment
Key Concepts
• Language Server Protocol (LSP) triggers• Tree-sitter AST syntax chunking• Two-tier model routing (fast inline vs deep reasoning)• Background compiler syntax validation
Principal

Autonomous Software Engineering Agent & Sandbox Harness (Devin / Claude Computer Use)

Orchestrate an autonomous agent that explores codebases, edits files, executes test suites, and iterates until builds pass.

Asked at
Cognition (Devin)AnthropicOpenAIMeta
Key Concepts
• ReAct plan-act-observe loops• Step-budget clamps & runaway prevention• MicroVM execution sandboxes (Firecracker / gVisor)• State checkpoint event logs
Staff

Real-Time Grounded AI Search & Answer Engine (Perplexity / SearchGPT)

Decompose user queries into parallel web searches, scrape candidate documents, rerank snippets, and stream answers with verifiable citations.

Asked at
PerplexityOpenAI (SearchGPT)GoogleMicrosoft (Bing Copilot)
Key Concepts
• Query decomposition & fan-out• Parallel web scraping & HTML text extraction• Cross-encoder passage reranking• Streaming inline markdown citations• Hallucination verification agents
Senior

Universal AI Gateway & Semantic Cache (LiteLLM / Cloudflare AI Gateway)

Provide a unified multi-provider control plane with vector semantic caching, automated fallbacks, dynamic model routing, and token quotas.

Asked at
CloudflarePortkeyScale AIDatadog
Key Concepts
• Vector cosine similarity semantic caching (Redis)• Adaptive cost & latency model routing• Provider circuit breaking & failover• Sliding-window token bucket rate limits• PII redaction
Senior

Automated LLM Evaluation Platform & LLM-as-a-Judge (LangSmith / Braintrust)

Run regression test suites, track prompt drift, and score generated outputs against golden datasets using automated judge models.

Asked at
BraintrustScale AIDatadogAnthropic
Key Concepts
• Immutable golden benchmark datasets• Chain-of-Thought judge models• Faithfulness & answer relevance scoring• CI/CD automated regression gates
Staff

Full-Duplex Real-Time Voice Agent & Interruption Engine (OpenAI Realtime / Gemini Live)

Stream bidirectional audio for natural voice conversations with client-side voice activity detection and instant barge-in cancellation.

Asked at
OpenAIGoogle DeepMindElevenLabsDeepgram
Key Concepts
• Bidirectional WebSocket & WebRTC Opus transport• Voice Activity Detection (VAD) barge-in interruption• Abort signal token generation cancellation• Speech-to-speech native inference
Staff

Distributed Text-to-Image & Video Generation Pipeline (Midjourney / Sora / Flux)

Process high-volume creative prompts with safety classifiers, tiered GPU priority scheduling, and latent diffusion multi-step denoising.

Asked at
MidjourneyRunwayOpenAIBlack Forest Labs (Flux)
Key Concepts
• Prompt safety & aesthetic expansion• Tiered GPU priority job queues• Latent diffusion multi-step denoising (FP8/INT8)• Progressive CDN preview delivery
Senior

Long-Term Conversational AI Memory & Persona Store (ChatGPT Memory / Mem0)

Extract declarative facts, user preferences, and entity relationship graphs across chat sessions and inject relevant memories on demand.

Asked at
OpenAICharacter.aiAppleMeta
Key Concepts
• Asynchronous declarative fact extraction worker• Entity-attribute relational store + dense vector embeddings• Top-k relevance re-ranking for system prompt injection• User memory deletion & privacy controls
Senior

Real-Time AI Guardrails & Prompt Injection Firewall (NeMo Guardrails / Llama Guard)

Inspect user prompts and model responses to block direct jailbreaks, indirect injections, and confidential credential leaks.

Asked at
CloudflarePalo Alto NetworksMicrosoftGoogle
Key Concepts
• Dual-phase pre-prompt & post-generation inspection• Jailbreak & prompt injection classifier heuristics• Sensitive entity detection (NER)• Surrogate token vaulting & de-anonymization
Principal

Distributed LLM Training & Fault-Tolerant Checkpointing System (Megatron-LM / DeepSpeed / FSDP)

Orchestrate multi-node GPU training across thousands of accelerators with 3D parallelism and non-blocking asynchronous checkpoint saves.

Asked at
Meta (Llama)Google DeepMindxAICoreWeave
Key Concepts
• 3D parallelism (Tensor, Pipeline, Data / ZeRO-3)• Asynchronous non-blocking checkpoint dumping to distributed NVMe/S3• InfiniBand & RoCE collective communication (NCCL)• Automated worker failure recovery & health checks
Senior

Synthetic Data Generation & RLHF / DPO Alignment Pipeline

Generate millions of high-quality synthetic training pairs, score model outputs, and execute direct preference optimization alignment.

Asked at
AnthropicScale AIDatabricksOpenAI
Key Concepts
• Self-instruct prompt generation & rejection sampling• Pairwise preference dataset synthesis• Direct Preference Optimization (DPO) training loops• Automated critique & revision loops
Senior

Enterprise Multi-Modal Document Intelligence & Table Parsing Engine

Ingest scanned enterprise PDFs, slides, and invoices with vision-based layout analysis, table reconstruction, and hierarchical chunking.

Asked at
DatabricksGleanAmazon BedrockGoogle Cloud
Key Concepts
• Vision-based bounding box layout detection• Structured table markdown reconstruction• Hierarchical parent-child chunking• Multi-modal vector indexing
Staff

High-Concurrency Sandboxed Code Execution Platform for AI (E2B / Modal / Firecracker)

Run untrusted Python and bash code generated by AI agents inside disposable, secured microVM environments.

Asked at
ModalE2BReplitCloudflare Workers
Key Concepts
• MicroVM rapid cold starts• Copy-on-write root filesystems• Strict system call filtering (seccomp / eBPF)• Network egress isolation & strict resource quotas
Staff

Speculative Decoding & Draft Model Inference Acceleration (Medusa / vLLM)

Accelerate large model token generation by employing a lightweight draft model to speculate multiple candidate tokens in parallel.

Asked at
Together AIGroqAnyscaleGoogle DeepMind
Key Concepts
• Lightweight draft model speculation• Target model parallel verification forward pass• Tree-based attention verification kernels• Dynamic speculative length adaptation
Senior

Scalable Function Calling & Agent Tool Execution Broker

Manage thousands of OpenAPI tool definitions, dynamically filter context schemas, and safely execute distributed external actions.

Asked at
OpenAILangChainAnthropicStripe
Key Concepts
• OpenAPI tool definition registries• Dynamic semantic tool schema filtering• Idempotent tool execution & token leases• Asynchronous polling for long-running workflows• Compensating rollback transactions
Staff

Enterprise Hybrid RAG Engine with Reciprocal Rank Fusion & GraphRAG

Combine dense vector similarity, sparse keyword search, and knowledge graph entity traversal to eliminate RAG retrieval blind spots.

Asked at
MicrosoftNeo4jCohereDatabricks
Key Concepts
• Knowledge graph entity & relationship extraction• Dense vector + sparse BM25 fusion (RRF)• Multi-hop reasoning retrieval paths• Contextual document compression
Senior

Multi-Turn AI Customer Support & Action Execution Agent (Klarna / Intercom Fin)

Resolve customer issues autonomously with intent classification, policy-bounded tool execution, and seamless human agent handoffs.

Asked at
KlarnaIntercomSalesforceZendesk
Key Concepts
• Intent triage & customer sentiment tracking• Policy-bounded tool execution with guardrails• Live conversation summarization• Graceful human escalation handoff
Principal

Vision-Language Action Agent & Desktop Automation (Claude Computer Use / Adept)

Control graphical desktop interfaces by taking screenshots, detecting interactive UI elements, and generating mouse and keyboard events.

Asked at
AnthropicAdeptMicrosoftOpenAI
Key Concepts
• Screen capture coordinate scaling• Vision transformer UI element localization• Mouse & keyboard action sequencing• Visual state verification feedback loops
Senior

High-Throughput Web Scraping & Synthetic Data Purification Pipeline for Pre-Training (Common Crawl / FineWeb)

Process petabyte-scale raw web crawls through fuzzy deduplication, quality filtering, PII removal, and synthetic document refinement.

Asked at
Hugging FaceMetaMistralxAI
Key Concepts
• MinHash LSH fuzzy text deduplication• FastText heuristic quality & language classification• Petabyte-scale distributed PII stripping• Synthetic document rephrasing & toxicity removal
The 6-Step Blueprint

Six areas to cover in your explanation.

Use these areas to organize your practice. Adjust the depth to the role and the time available.

  1. 01

    Clarify Requirements & Scope

    Lock down functional APIs and non-functional SLAs (availability, latency, consistency).

  2. 02

    Capacity & Resource Math

    Derive peak QPS, storage growth per year, cache memory, and network ingress/egress.

  3. 03

    High-Level Component Topology

    Draw client gateways, load balancers, microservices, caches, and storage tiers.

  4. 04

    Database & Data Model Schema

    Define relational vs NoSQL schemas, primary keys, partition strategies, and indexes.

  5. 05

    Deep-Dive Bottlenecks & SPOFs

    Address cache invalidation, network partitions, replication lag, and leader elections.

  6. 06

    Trade-Off Defense

    Defend architectural compromises (CAP theorem, sync vs async, consistency vs latency).

FAQ

System design questions

How many system design questions are included in ClawPad's catalog?

The question bank contains 100 system design practice questions across 15 architectural domains. This page also links to 20 worked blueprints with example designs.

How are ClawPad system design blueprints structured?

The worked blueprints cover requirements, capacity estimates, architecture components, data models, and trade-offs. Treat their numbers as example assumptions and explain how your design would change if those assumptions changed.

How can I rehearse a system design interview?

Choose a question and spend 35 minutes explaining requirements, estimates, data flow, a failure case, and a tradeoff without a reference. Use the final ten minutes to review the evidence and retry the weakest section. ClawPad can provide structured answers and diagrams for study.

Structured answer study

Study a guided answer, then explain your own design.

Download ClawPad