System Design Practice Problems

Each problem follows the same structure, so you can practise the interview framework every time:

Problem → Clarifying questions → Requirements → Estimates → Architecture diagram → Data model → Deep dives → Trade-offs → Failure modes → Senior signals → Follow-ups → Rubric

How to practise: read only the Problem section, set a 45-minute timer, design on paper while talking out loud, then compare with the reference answer and tick the rubric. Re-do anything under 70% a few days later.

Core data platform designs

#	Problem	Difficulty	Key concepts
1	Real-time clickstream analytics	Medium	Kafka, dedup, sessionization, late data, OLAP
2	Real-time payment fraud detection	Hard	Sync vs async paths, feature store, latency budget
3	Ad click aggregation and billing	Hard	Exactly-once, approximate vs exact paths, skew
4	CDC from 200 OLTP tables into a lakehouse	Medium	Debezium, MERGE, SCD2, schema evolution
5	Real-time top-K trending	Medium	Bucketed windows, heavy hitters, count-min sketch
6	Ride-hailing surge pricing data	Hard	Geospatial (H3), keyed state, timers, safety
7	Connected-vehicle telemetry	Hard	MQTT, time series, edge filtering, privacy
8	Feature platform for recommendations	Hard	Point-in-time joins, skew, online serving
9	A/B testing data pipeline	Hard	Exposure, sufficient stats, SRM, CUPED
10	Auditable financial reporting	Medium	Ledger modelling, reconciliation, SOX

Platform, governance and AI

#	Problem	Difficulty	Key concepts
11	Governed lakehouse for 8 domains	Hard	Catalog layout, ABAC, shared standards, CI/CD
12	Data quality and observability platform	Medium	Rules as code, anomaly detection, WAP, routing
13	GDPR right-to-erasure platform	Hard	Discovery via tags, VACUUM vs SLA, crypto-shredding
14	Legacy warehouse migration with AI-assisted SQL conversion	Hard	Validation harness, LLM agents, parallel runs
15	Enterprise RAG knowledge assistant	Hard	Incremental embedding, ACL-aware retrieval, evals
16	LLM observability and evaluation platform	Medium	Tracing, cost attribution, judge calibration

Product and operations data systems

#	Problem	Difficulty	Key concepts
17	Usage metering and billing	Hard	Effectively-once metering, as-of pricing, ledger, finalisation, reconciliation
18	Real-time inventory availability	Hard	Ledger + derived state, sequence dedup, reservations, oversell prevention
19	Logs and metrics observability platform	Hard	Tiered storage, cardinality, sampling, reliable alerting, cost
20	Real-time customer data platform	Hard	Identity graph, streaming segments, consent-aware activation

Scenario and debugging questions (30 min)

Problem	Focus
Data reconciliation between conflicting sources	Data quality, conflict resolution
Exactly-once in a Kafka payment pipeline	Idempotent producers, transactions, sinks
Pipeline fails only on Mondays	Structured debugging

Suggested order

Week 1: 1 → 4 → 5 → 10 (fundamentals: streaming, CDC, windows, correctness)
Week 2: 3 → 2 → 8 → 11 (exactly-once, low latency, ML data, platform)
Week 3: 7 → 6 → 9 → 13 (domain-specific depth)
Week 4: 17 → 18 → 20 → 19 (product data systems: money, inventory, identity, observability)
AI track: 15 → 16 → 14