The Complete System Design Interview Question Bank: 50 Questions by Category

Most system design interview prep resources give you a list of 15 questions and call it a day. The problem: interviewers have gotten wise. Candidates who memorized "Design a URL shortener" deliver polished but shallow answers. The signal disappears.

This question bank is different. It contains 50 questions organized across 10 categories, each with a difficulty rating, specific evaluation criteria, and a follow-up probe question. Whether you are an interviewer building a question rotation or a candidate preparing for your next loop, this is designed to be the resource you actually use, not just bookmark.

If you are new to system design interviews entirely, start with our complete guide to system design interviews first. If you already know the format and want to understand how to score answers, read our rubric-based evaluation framework.

How to Use This Question Bank

Before jumping into the questions, a quick calibration guide. Not every question works for every level. Using a Staff-level question on a junior candidate wastes everyone's time. Using a Junior-level question on a Staff candidate gives you zero signal.

Junior (0-3 years): Expect clear requirements gathering, a reasonable HLD with 3-5 components, and at least one explicit tradeoff discussion. Do not penalize for missing edge cases or advanced scaling patterns. Focus on structured thinking.

Senior (4-8 years): Expect thorough requirements, a detailed HLD with clear data flow, LLD for at least one critical component, multiple tradeoff analyses with justification, and proactive discussion of failure modes. They should drive the conversation, not wait for prompts.

Staff (8+ years): Expect everything above, plus cross-system thinking (how does this interact with adjacent systems?), organizational implications (which team owns what?), migration strategies (how do you get from here to there without downtime?), and cost/operational tradeoff reasoning. Staff candidates should reframe the problem, not just solve it.

Dynamic constraints: For every category below, I have included a note on mid-interview constraint injection. This is the single best technique for separating prepared answers from actual thinking. Inject a new requirement 20 minutes in and watch how the candidate adapts. Do they panic and start over? Do they bolt it on as an afterthought? Or do they systematically evaluate the impact on their existing design? That response tells you more than the first 20 minutes combined.

---

1. E-Commerce and Marketplace

These questions test multi-stakeholder reasoning (buyers, sellers, platform), inventory/state management, and transaction integrity. Strong answers balance user experience with data consistency.

Dynamic constraint to inject mid-interview: "The CEO just told you we are expanding to 3 new countries next quarter. Each has different tax rules, currencies, and regulatory requirements. How does your design change?"

Q1. Design an e-commerce product search and browse system like Amazon's catalog

Difficulty: Senior

Key evaluation areas:

• Indexing strategy for millions of SKUs (inverted index, faceted search, relevance ranking)
• How search results stay fresh as inventory and pricing change in real time
• Personalization layer: balancing recommendation relevance with search intent

Follow-up: "A seller lists 50,000 new products per day. How do you ensure search freshness without reindexing the entire catalog?"

---

Q2. Design a shopping cart that works across devices and survives crashes

Difficulty: Junior

Key evaluation areas:

• Server-side vs client-side cart state (and the merge strategy when a user logs in on a new device)
• Handling inventory reservation and release (what happens when a user abandons a cart?)
• Session management and guest-to-authenticated cart migration

Follow-up: "Two tabs are open. The user adds an item in Tab A and removes it in Tab B simultaneously. What happens?"

---

Q3. Design a real-time auction system like eBay

Difficulty: Staff

Key evaluation areas:

• Bid ordering guarantees under high concurrency (clock skew, distributed consensus)
• Anti-sniping mechanisms and their impact on system load patterns
• Cross-cutting concerns: fraud detection, bid retraction, seller trust scoring

Follow-up: "A celebrity tweets about an item and bid volume spikes 100x in 30 seconds. Walk me through what happens in your system."

---

Q4. Design the seller onboarding and verification pipeline for a marketplace

Difficulty: Senior

Key evaluation areas:

• Multi-step verification workflow (identity, bank account, product compliance) with state machine design
• Handling partial completions, retries, and manual review queues
• Data privacy: storing sensitive documents (PII, financial records) with appropriate access controls

Follow-up: "Regulators now require re-verification every 90 days. How do you retrofit this into your existing pipeline without breaking the flow for active sellers?"

---

Q5. Design a dynamic pricing engine for ride-sharing or hotel bookings

Difficulty: Staff

Key evaluation areas:

• Real-time demand signal ingestion and price computation latency requirements
• Fairness constraints: price caps, anti-gouging rules, geographic sensitivity
• Feedback loops: how pricing changes affect demand, and how to prevent oscillation

Follow-up: "Your pricing algorithm sets a 4x surge price and it goes viral on social media as price gouging. What systemic guardrails should have prevented this?"

---

2. Social Media and Communication

These questions test real-time delivery, fan-out patterns, content moderation, and privacy-aware data modeling. The best answers distinguish between read-heavy and write-heavy paths.

Dynamic constraint to inject mid-interview: "We just acquired a company with 200M users. How do you merge their social graph into yours without breaking either product?"

Q6. Design a news feed system like Twitter/X's home timeline

Difficulty: Senior

Key evaluation areas:

• Fan-out-on-write vs fan-out-on-read (and the hybrid approach for high-follower accounts)
• Ranking and relevance scoring beyond chronological order
• Handling deletes, edits, and content takedowns that must propagate to already-rendered feeds

Follow-up: "An account with 50 million followers posts. Walk me through the exact sequence of events from post creation to feed delivery."

---

Q7. Design a group chat system supporting up to 100,000 members

Difficulty: Senior

Key evaluation areas:

• Message delivery guarantees (at-least-once vs exactly-once) and ordering semantics
• Read receipts at scale (you cannot fan out receipts to 100K members per message)
• Storage strategy: hot messages in memory, warm in fast storage, cold archived

Follow-up: "A government mandates that all messages must be stored for 7 years with full audit trail, but users expect end-to-end encryption. How do you handle this?"

---

Q8. Design a content moderation pipeline for user-generated text, images, and video

Difficulty: Staff

Key evaluation areas:

• Multi-stage pipeline: automated ML classifiers, confidence thresholds, human review queues
• Latency requirements: pre-publish blocking vs post-publish takedown and the UX tradeoffs
• Appeals workflow, audit logging, and false positive rates as operational metrics

Follow-up: "Your automated system has a 2% false positive rate. At 10M posts/day, that is 200K wrongly removed posts. How do you handle this operationally?"

---

Q9. Design a stories/ephemeral content system (24-hour disappearing posts)

Difficulty: Junior

Key evaluation areas:

• TTL-based storage and cleanup strategy (do you actually delete or soft-delete?)
• View tracking and viewer lists with privacy controls
• CDN strategy for media-heavy content with a known expiration window

Follow-up: "A user screenshots a story and the poster wants to be notified. How do you detect this across iOS, Android, and web?"

---

Q10. Design a notification system that handles push, email, SMS, and in-app

Difficulty: Senior

Key evaluation areas:

• Preference management: per-channel, per-category, quiet hours, frequency caps
• Delivery pipeline: templating, channel routing, retry logic, delivery confirmation
• Deduplication: ensuring a user does not receive the same notification across multiple channels

Follow-up: "You send a flash sale notification to 10M users. Email provider rate-limits you to 100K/hour. How do you prioritize who gets notified first?"

---

3. Streaming and Media

These questions test CDN architecture, encoding pipelines, adaptive delivery, and the unique challenges of time-sensitive content. Strong answers reason about bandwidth costs.

Dynamic constraint to inject mid-interview: "We need to add live captioning in 40 languages with under 3 seconds of delay. Where does this fit in your pipeline?"

Q11. Design a video streaming service like Netflix

Difficulty: Senior

Key evaluation areas:

• Transcoding pipeline: adaptive bitrate encoding, format variants, and when to kick off encoding
• CDN architecture: edge caching strategy, origin shield, cache invalidation for content updates
• Client-side adaptive bitrate switching based on network conditions

Follow-up: "Your most popular show drops at midnight and 30M users hit play within 60 seconds. What breaks first in your architecture?"

---

Q12. Design a live streaming platform (Twitch-style) with real-time chat

Difficulty: Staff

Key evaluation areas:

• Ingest → transcode → distribute latency budget (sub-5-second glass-to-glass for interactive streams)
• Chat at scale: message ordering, spam filtering, and emote rendering with 500K concurrent viewers
• Streamer tools: stream health monitoring, clip creation, VOD recording running in parallel

Follow-up: "A streamer's internet connection drops for 10 seconds then reconnects. What does the viewer see? What does the streamer see? What does the system do?"

---

Q13. Design a music recommendation engine like Spotify Discover Weekly

Difficulty: Senior

Key evaluation areas:

• Collaborative filtering vs content-based approaches (and why you need both)
• Cold start problem: recommending to new users with no listening history
• Feedback signals: explicit (likes, skips) vs implicit (listen duration, repeat plays)

Follow-up: "A user listens to death metal 99% of the time but played a children's lullaby playlist once for their kid. How does your model avoid corrupting their recommendations?"

---

Q14. Design a podcast distribution and playback system

Difficulty: Junior

Key evaluation areas:

• RSS feed ingestion, audio file storage, and CDN distribution
• Playback state sync across devices (resume where you left off)
• Download management for offline listening: storage quotas, background downloads, expiration

Follow-up: "A podcast publisher pushes an update to an episode (correcting an error). How do users who already downloaded the old version get notified?"

---

Q15. Design an AI-powered video clip extraction tool (auto-highlight reels)

Difficulty: Staff

Key evaluation areas:

• ML pipeline for identifying "interesting" segments (applause detection, sentiment spikes, visual activity)
• Inference cost management: batch processing vs real-time, GPU scheduling
• User feedback loop to improve clip quality over time

Follow-up: "Processing a 2-hour video takes 45 minutes on your current GPU fleet. A creator wants clips within 5 minutes of stream end. What is your approach?"

---

4. Search and Discovery

These questions test indexing, ranking, relevance, and the feedback loops between user behavior and search quality. The best answers discuss measurement (how do you know your search is good?).

Dynamic constraint to inject mid-interview: "We now need to support voice search and image search in addition to text. How does your architecture accommodate multi-modal queries?"

Q16. Design a web search engine's crawling and indexing pipeline

Difficulty: Staff

Key evaluation areas:

• Crawl scheduling: politeness (robots.txt), freshness requirements, priority queuing
• Distributed indexing: sharding strategy, index compaction, and serving latency requirements
• Handling adversarial content: SEO spam, cloaking, link farms

Follow-up: "A major news event happens and 10,000 new articles publish in an hour. How does your crawler detect and prioritize these for immediate indexing?"

---

Q17. Design an autocomplete/typeahead search system

Difficulty: Junior

Key evaluation areas:

• Trie or prefix-based data structure for fast prefix lookups
• Ranking suggestions by popularity, recency, and personalization
• Latency budget: under 100ms for typeahead to feel instant

Follow-up: "A trending topic spikes in searches. How quickly do your autocomplete suggestions reflect this, and what is the update mechanism?"

---

Q18. Design a recommendation system for an e-commerce homepage

Difficulty: Senior

Key evaluation areas:

• Feature engineering from browsing behavior, purchase history, and demographic data
• Real-time vs batch recommendation generation and when to use each
• A/B testing infrastructure for evaluating recommendation algorithm changes

Follow-up: "Your recommendation model has a bug that promotes expensive items regardless of relevance. It has been running for 3 days before anyone notices. How do you detect and recover?"

---

Q19. Design a plagiarism detection system for academic submissions

Difficulty: Senior

Key evaluation areas:

• Document fingerprinting: n-gram hashing, MinHash/LSH for approximate matching
• Corpus management: indexing millions of existing documents for comparison
• AI-generated text detection as an additional signal alongside traditional plagiarism

Follow-up: "Students start using AI to paraphrase plagiarized content, defeating your n-gram matching. How does your system evolve?"

---

Q20. Design a semantic search system using vector embeddings

Difficulty: Staff

Key evaluation areas:

• Embedding generation pipeline: model selection, batch vs real-time, versioning when model changes
• Vector database selection and indexing (HNSW, IVF) with latency/recall tradeoffs
• Hybrid search: combining vector similarity with keyword matching and metadata filtering

Follow-up: "You upgrade your embedding model and recall improves 15%, but every existing document needs re-embedding. You have 500M documents. What is your migration plan?"

---

5. Payments and FinTech

These questions demand precise reasoning about consistency, auditability, and failure handling. Handwaving is unacceptable here. Strong answers cite specific consistency guarantees.

Dynamic constraint to inject mid-interview: "A regulator requires that all transactions over $10,000 must be held for 24-hour review before settlement. Insert this into your flow."

Q21. Design a peer-to-peer payment system like Venmo

Difficulty: Senior

Key evaluation areas:

• Double-entry bookkeeping model and how transactions are represented in the database
• Idempotency: handling duplicate payment requests from retries without charging twice
• Settlement flow: instant balance vs ACH settlement timing and the liability implications

Follow-up: "A user disputes a payment 30 days later. Walk me through the data you need to resolve this and the reversal flow."

---

Q22. Design a fraud detection system for credit card transactions

Difficulty: Staff

Key evaluation areas:

• Real-time scoring pipeline: feature extraction, model inference, and decision within 100ms
• Balancing false positive rate (blocking legitimate purchases) vs false negative rate (letting fraud through)
• Model retraining pipeline: how quickly can you adapt to new fraud patterns?

Follow-up: "A new fraud pattern emerges on a Friday night. Your ML team is not available until Monday. What automated defenses does your system have?"

---

Q23. Design a subscription billing system with plan changes, proration, and invoicing

Difficulty: Senior

Key evaluation areas:

• State machine for subscription lifecycle (trial, active, past_due, canceled, paused)
• Proration logic when users upgrade or downgrade mid-billing-cycle
• Invoice generation, tax calculation, and payment retry scheduling for failed charges

Follow-up: "A customer downgrades from annual to monthly plan with 7 months remaining. How do you handle the credit, and what does the invoice look like?"

---

Q24. Design a multi-currency wallet system for a global fintech app

Difficulty: Staff

Key evaluation areas:

• Exchange rate management: sourcing, caching, and staleness tolerances for display vs execution
• Ledger design: storing amounts in minor units, handling rounding across conversions
• Regulatory compliance: KYC/AML per jurisdiction, transaction limits, reporting requirements

Follow-up: "A user holds USD, converts to EUR, then converts back to USD, and the amount is different due to spread. Three users file complaints. How does your system explain and prevent confusion?"

---

Q25. Design a payment gateway integration layer that supports multiple providers (Stripe, Adyen, PayPal)

Difficulty: Junior

Key evaluation areas:

• Adapter pattern: abstracting provider-specific APIs behind a common interface
• Failover strategy: routing to a backup provider when the primary is down
• Webhook handling: idempotent processing of payment status callbacks

Follow-up: "Stripe is down for 15 minutes. You failed over to Adyen. Stripe comes back and starts sending delayed webhooks for transactions you already retried on Adyen. How do you prevent double-charging?"

---

6. Cloud Infrastructure

These questions test distributed systems fundamentals. Expect candidates to reason about consensus, partitioning, replication, and operational concerns. Weak answers regurgitate CAP theorem without understanding it.

Dynamic constraint to inject mid-interview: "Your largest customer just told you they require data residency in the EU. No data can leave eu-west-1. How does this affect your design?"

Q26. Design a distributed key-value store like Redis or DynamoDB

Difficulty: Staff

Key evaluation areas:

• Partitioning strategy (consistent hashing) and replication for fault tolerance
• Consistency model: tunable consistency (W+R>N) with clear explanation of the tradeoffs
• Failure handling: node failures, network partitions, and how the system recovers

Follow-up: "A network partition splits your cluster into two halves. Both halves continue accepting writes. The partition heals. What happens to conflicting writes?"

---

Q27. Design a container orchestration system's scheduling component

Difficulty: Staff

Key evaluation areas:

• Resource modeling: CPU, memory, GPU, and custom resource types with bin-packing optimization
• Scheduling constraints: affinity, anti-affinity, taints, tolerations, topology spread
• Preemption logic: when and how to evict lower-priority workloads

Follow-up: "A high-priority job needs 8 GPUs but they are spread across 4 nodes with 2 each. Your scheduler requires co-located GPUs. What are your options?"

---

Q28. Design a log aggregation and search system like the ELK stack

Difficulty: Senior

Key evaluation areas:

• Ingestion pipeline: handling bursty log volume without backpressure collapsing the pipeline
• Storage tiering: hot (recent, fast queries), warm (older, slower), cold (archived, cheap)
• Query performance: full-text search over terabytes of logs with sub-second response

Follow-up: "A misconfigured service starts emitting 10x its normal log volume. How does your system protect itself from this single noisy tenant?"

---

Q29. Design a feature flag and configuration management system

Difficulty: Junior

Key evaluation areas:

• Flag evaluation: local SDK with cached rules vs remote API calls per evaluation
• Rollout strategies: percentage-based, user-segment-based, canary, kill switch
• Audit trail: who changed what flag when, and the ability to instant-rollback

Follow-up: "A flag change causes a production incident. The engineer who made the change is on a flight. How does your system enable anyone on the team to identify and revert the problematic change?"

---

Q30. Design a multi-region active-active database deployment

Difficulty: Staff

Key evaluation areas:

• Conflict resolution strategy for concurrent writes to the same record in different regions
• Latency implications of cross-region replication and how to minimize user-visible impact
• Failover mechanics: how traffic shifts when a region goes down, DNS vs application-level routing

Follow-up: "You are active-active in US-East and EU-West. A submarine cable cut increases cross-region latency from 80ms to 800ms. What degrades gracefully and what breaks?"

---

7. Real-Time Systems

These questions test latency sensitivity, event ordering, and the gap between "real-time" as a marketing term and real-time as an engineering constraint.

Dynamic constraint to inject mid-interview: "Regulations now require a complete audit trail of every state change with tamper-proof timestamps. How do you add this without affecting your latency budget?"

Q31. Design a real-time collaborative document editor like Google Docs

Difficulty: Staff

Key evaluation areas:

• Conflict resolution: OT (Operational Transformation) vs CRDT with honest assessment of complexity tradeoffs
• Cursor and presence tracking for 50+ simultaneous editors
• Offline editing and sync: how edits made offline merge when connectivity returns

Follow-up: "Two users are editing the same paragraph simultaneously. One deletes a sentence that the other is inserting text into. What does each user see?"

---

Q32. Design a ride-matching system for a ride-sharing app

Difficulty: Senior

Key evaluation areas:

• Geospatial indexing for efficient proximity queries (geohash, R-tree, H3)
• Matching algorithm: balancing rider wait time, driver utilization, and trip profitability
• Handling supply/demand imbalance: surge pricing triggers, driver repositioning suggestions

Follow-up: "A concert ends and 5,000 ride requests appear in a 2-block radius within 3 minutes. Your matching algorithm normally handles 50 requests/second for this zone. What happens?"

---

Q33. Design a stock trading order book and matching engine

Difficulty: Staff

Key evaluation areas:

• Order matching: price-time priority, order types (market, limit, stop), partial fills
• Latency requirements: sub-millisecond matching with deterministic execution
• Market data distribution: broadcasting price updates to thousands of subscribers with minimal skew

Follow-up: "Your matching engine crashes and restarts. How do you rebuild the order book state and guarantee no orders were lost or duplicated?"

---

Q34. Design a multiplayer game server for a battle royale game (100 players)

Difficulty: Senior

Key evaluation areas:

• State synchronization: authoritative server model, tick rate, client-side prediction, lag compensation
• Network optimization: UDP over TCP, delta compression, interest management (only send nearby player data)
• Cheat prevention: server-side validation of all game-critical actions

Follow-up: "A player has 200ms latency while others have 20ms. On their screen they dodged a bullet, but on the server they did not. How do you handle this?"

---

Q35. Design an IoT sensor data collection and alerting system for a factory floor

Difficulty: Senior

Key evaluation areas:

• Data ingestion: handling thousands of sensors reporting every second with potential connectivity gaps
• Edge vs cloud processing: which alerts need sub-second response (fire, leak) vs which can tolerate minutes
• Time-series storage: efficient compression, retention policies, downsampling for historical queries

Follow-up: "A sensor starts reporting values that are technically within normal range but the rate of change is anomalous. How does your system detect this pattern?"

---

8. Data Pipelines

These questions test batch and stream processing, data quality, and the practical challenges of moving data through systems reliably.

Dynamic constraint to inject mid-interview: "GDPR requires that when a user requests deletion, their data must be purged from all pipeline stages, including intermediate storage and derived datasets, within 30 days. How does your pipeline support this?"

Q36. Design a real-time analytics dashboard for a SaaS product

Difficulty: Senior

Key evaluation areas:

• Event ingestion: handling spiky write volumes without data loss (Kafka/Kinesis, backpressure handling)
• Pre-aggregation strategy: what to compute at write time vs query time
• Dashboard query performance: sub-second response for time-range queries over billions of events

Follow-up: "Your dashboard shows 1.2M page views yesterday, but the marketing team's Google Analytics shows 1.4M. How do you debug the discrepancy?"

---

Q37. Design an ETL pipeline that syncs data between a transactional database and a data warehouse

Difficulty: Junior

Key evaluation areas:

• Change data capture (CDC) vs periodic full dumps and the tradeoffs of each
• Schema evolution: how the pipeline handles new columns, renamed fields, or type changes in the source
• Error handling: dead letter queues, retry logic, and alerting on pipeline failures

Follow-up: "A developer renames a column in the source database. Your ETL pipeline breaks at 2 AM. What does your automated recovery look like?"

---

Q38. Design a machine learning feature store

Difficulty: Staff

Key evaluation areas:

• Online (low-latency serving) vs offline (batch training) feature retrieval with consistency guarantees
• Feature versioning: ensuring training and serving use the same feature definitions
• Backfill strategy: recomputing features historically when definitions change

Follow-up: "A data scientist discovers that a feature used in production has a subtle bug that has been computing incorrect values for 2 weeks. What is your incident response?"

---

Q39. Design a data quality monitoring system

Difficulty: Senior

Key evaluation areas:

• Expectation framework: schema validation, statistical distribution checks, freshness monitoring
• Anomaly detection: distinguishing real data issues from legitimate distribution shifts
• Lineage tracking: when bad data is detected, tracing it back to the source

Follow-up: "Your monitoring flags a 40% drop in daily active users in the analytics table. Is it a data pipeline issue or an actual product problem? How does your system help distinguish the two?"

---

Q40. Design a CDC (change data capture) system for a microservices architecture

Difficulty: Senior

Key evaluation areas:

• Log-based CDC vs trigger-based vs polling and the tradeoffs for each approach
• Ordering guarantees: ensuring consumers see changes in the correct sequence
• Schema registry integration: how consumers handle upstream schema changes gracefully

Follow-up: "Service A publishes CDC events. Service B consumes them. Service A deploys a schema change that adds a required field. Service B has not been updated yet. What happens?"

---

9. API Design

These questions test interface thinking: how candidates design contracts between systems. The best answers reason about versioning, backward compatibility, and developer experience.

Dynamic constraint to inject mid-interview: "Your API now needs to support a mobile client on 2G networks in emerging markets. Average payload must be under 5KB. How do you adapt?"

Q41. Design a public REST API for a calendar application (Google Calendar-style)

Difficulty: Junior

Key evaluation areas:

• Resource modeling: events, calendars, attendees, and their relationships
• Recurring events: representation format (RRULE), exception handling, and timezone complexity
• Pagination and filtering for event list queries

Follow-up: "A recurring weekly meeting has 200 attendees. One instance needs to be rescheduled. How does your API model the difference between 'this instance' and 'all future instances'?"

---

Q42. Design a rate limiting system for a public API

Difficulty: Senior

Key evaluation areas:

• Algorithm selection: token bucket, sliding window, or leaky bucket with justification
• Distributed rate limiting: coordination across multiple API gateway instances
• Response design: appropriate headers (X-RateLimit-Remaining, Retry-After) and status codes

Follow-up: "A customer uses 5 different API keys across 5 microservices but should share a single rate limit. How does your system handle per-organization limits?"

---

Q43. Design a GraphQL API gateway that federates across multiple backend services

Difficulty: Staff

Key evaluation areas:

• Schema composition: how services register their schema fragments and resolve cross-service references
• Query planning: optimizing execution to minimize backend round-trips
• Security: query depth limiting, cost analysis, and preventing abuse through expensive nested queries

Follow-up: "Service A owns the User type and Service B extends it with a orders field. Service A deploys a breaking change to User. How does your gateway handle partial federation failures?"

---

Q44. Design a webhook delivery system with guaranteed at-least-once delivery

Difficulty: Senior

Key evaluation areas:

• Delivery pipeline: persistent queue, retry schedule with exponential backoff, dead letter after max retries
• Receiver verification: HMAC signatures, challenge-response registration
• Observability: delivery logs, failure reasons, and a self-service dashboard for consumers to debug

Follow-up: "A consumer's endpoint is returning 200 OK but silently dropping payloads (they have a bug). They contact you claiming they are not receiving webhooks. How do you help them debug this?"

---

Q45. Design a versioning strategy for a REST API with thousands of active consumers

Difficulty: Senior

Key evaluation areas:

• Versioning approach: URL path vs header vs content negotiation with tradeoffs
• Backward compatibility: additive changes that do not break existing clients
• Deprecation lifecycle: sunset headers, migration guides, and usage monitoring for old versions

Follow-up: "You need to rename a field from userName to user_name across your API. 400 active integrations use the old name. What is your 6-month migration plan?"

---

10. Mobile and IoT

These questions test resource-constrained thinking: limited battery, unreliable networks, edge compute, and the unique challenges of deploying software to hardware you do not control.

Dynamic constraint to inject mid-interview: "Your device fleet just grew from 10,000 to 2 million units across 40 countries. Your current OTA update system pushes updates to all devices simultaneously. What breaks?"

Q46. Design an offline-first mobile application with eventual sync

Difficulty: Senior

Key evaluation areas:

• Local data model and conflict resolution strategy when syncing after extended offline periods
• Sync protocol: delta sync, version vectors, and bandwidth optimization for metered connections
• UX decisions: what does the user see during sync conflicts? Automatic resolution vs manual merge?

Follow-up: "A field technician has been working offline for 3 days. When they reconnect, 47 records conflict with changes made by other team members. What does the resolution flow look like?"

---

Q47. Design an OTA (over-the-air) firmware update system for 1 million IoT devices

Difficulty: Staff

Key evaluation areas:

• Staged rollout: canary to 1%, monitor, expand progressively with automatic rollback triggers
• Update mechanics: differential updates, A/B partition schemes for safe rollback
• Fleet management: tracking update status across heterogeneous device types and firmware versions

Follow-up: "A firmware update bricks 0.3% of devices due to a rare hardware variant you did not test. The devices are in remote locations. What is your recovery plan?"

---

Q48. Design a mobile push notification system that maximizes delivery rate while respecting battery life

Difficulty: Junior

Key evaluation areas:

• Platform differences: APNs (Apple) vs FCM (Google) and their distinct delivery guarantees
• Batching and coalescing: grouping notifications to reduce wake-ups
• Token management: handling token refresh, expired tokens, and uninstalls

Follow-up: "Your delivery rate on Android is 85% but on iOS it is 97%. What are the likely causes and how do you investigate?"

---

Q49. Design a smart home hub that coordinates 50+ IoT devices with sub-second response

Difficulty: Senior

Key evaluation areas:

• Protocol bridging: Zigbee, Z-Wave, Bluetooth, WiFi, Thread, and Matter compatibility
• Local processing vs cloud: which automations must run locally (safety-critical) vs can tolerate cloud roundtrip
• Scene orchestration: "Good night" triggers 15 device actions. How do you ensure all execute reliably?

Follow-up: "The internet goes down. Which automations continue working and which degrade? How does the user know the difference?"

---

Q50. Design a health monitoring wearable's data pipeline (continuous heart rate, SpO2, accelerometer)

Difficulty: Staff

Key evaluation areas:

• On-device processing: what to compute locally (step counting, arrhythmia detection) vs send to cloud
• Battery optimization: sampling rate adaptation based on activity state
• Clinical-grade data integrity: handling gaps, sensor noise, and calibration drift

Follow-up: "Your arrhythmia detection algorithm flags a false positive and the user receives a 'seek medical attention' alert. This has happened to 50 users this month. What is your engineering response?"

---

How to Evaluate Answers

Having 50 questions means nothing if you cannot consistently evaluate the responses. We built a 5-dimension rubric specifically for this purpose. It scores answers across:

1. Problem Framing (20%) — Did they clarify before designing?
2. System Decomposition (20%) — Did they break it into logical components?
3. Tradeoff Analysis (20%) — Did they justify their choices?
4. Scalability and Edge Cases (25%) — Did they think about what breaks?
5. User-Centric Design (15%) — Did they keep the end user in focus?

The rubric works whether you are evaluating live or reviewing recorded sessions. Each dimension is scored 1-5 with specific behavioral anchors at each level.

Difficulty Calibration: What to Expect at Each Level

The difficulty labels on each question above are starting points. What really differentiates levels is the depth and breadth of the answer, not the question itself. Here is what good looks like at each level for any question in this bank:

Junior candidates should:

• Spend 3-5 minutes on requirements gathering
• Produce a clear HLD with 3-5 well-defined components
• Identify at least one tradeoff and explain their choice
• Handle the basic happy path and one failure scenario
• Communicate their design clearly with a diagram

Senior candidates should:

• Drive the entire conversation with minimal prompting
• Produce both HLD and LLD for critical path components
• Discuss 3+ tradeoffs with real-world justification
• Proactively raise scalability concerns and failure modes
• Handle the dynamic constraint injection without starting over
• Discuss monitoring, alerting, and operational runbook concerns

Staff candidates should:

• Reframe the problem before solving it ("The real question here is...")
• Reason about organizational boundaries (who owns what, how teams coordinate)
• Discuss migration from current state to proposed state (not just greenfield)
• Address cost, operational complexity, and hiring implications of their design
• Think in terms of platform capabilities, not just point solutions
• Handle multiple constraint injections gracefully
• Identify risks the interviewer did not mention

AI-Infrastructure Questions: The New Category

You will notice several questions above touch on AI/ML infrastructure (vector search, inference cost, ML feature stores, AI-generated content detection). This is intentional. In 2026, candidates who cannot reason about embedding pipelines, model serving costs, and GPU scheduling are missing a growing part of the infrastructure landscape. If you are building an interview loop today, include at least one AI-infrastructure question for senior and above.

Front-End System Design: A Note

Front-end system design is emerging as its own interview category, distinct from the backend-focused questions in this bank. Questions like "Design a component library's versioning system" or "Design a micro-frontend architecture for a large e-commerce site" test different skills: rendering performance, state management, bundle optimization, and accessibility. We will publish a dedicated front-end system design question bank separately.

What This Bank Does Not Cover

This bank focuses on system design for product and infrastructure engineers. It does not cover:

• Coding interviews (algorithm and data structure problems)
• Behavioral interviews (leadership, conflict resolution)
• Machine learning system design (model architecture, training pipeline design) — though several questions touch on ML infrastructure

For a broader understanding of where system design fits in the interview process, see our complete guide to system design interviews.

---

Build Your Interview Loop Faster

This question bank has 50 questions. AssessAI has 275+, organized across 20 categories, with AI-powered evaluation that scores candidate responses across all 5 dimensions automatically. No more interviewer calibration meetings. No more inconsistent scoring.

Paste a job description, get tailored questions for that specific role, and receive detailed scorecards with dimension-by-dimension breakdowns.

Try it free →