System Design Interview
Real-time fraud detection with ML inference, feature store, and event streaming.
Start with a single service applying rules to payments.
What was missing: No separation of concerns, no redundancy, no async, no observability.
Why that's risky: Single point of failure, no protection from spikes, no monitoring.
What gets added: Nothing yet (MVP).
Trade-offs: Simple, but not production ready.
Introduce edge, API gateway, and rate limiting for security and performance.
What was missing: No edge security, no rate limiting, no separation of gateway and rules.
Why that's risky: Vulnerable to DDoS, no traffic shaping, gateway logic not isolated.
What gets added: Edge, API gateway, and rate limiter.
Trade-offs: Slightly more complex.
Add load balancer and multiple rule engines for scale and redundancy.
What was missing: No redundancy, no horizontal scaling, single rule engine bottleneck.
Why that's risky: Single point of failure, cannot handle spikes.
What gets added: Load balancer, multiple rule engines.
Trade-offs: More complex deployment.
Introduce ML inference and feature store for advanced scoring.
What was missing: No ML scoring, no feature store, only rules.
Why that's risky: Rules alone miss subtle fraud, no learning from new patterns.
What gets added: ML inference, feature store.
Trade-offs: More moving parts, more operational complexity.
Add streaming analytics, async queue, worker pool, and monitoring for observability and feedback.
What was missing: No async analytics, no monitoring, no idempotency, no feedback loop.
Why that's risky: No observability, analytics can block fraud scoring, no alerting.
What gets added: Async queue, worker pool, monitoring/logging, idempotency.
Trade-offs: More moving parts, eventual consistency for analytics.
Combine rules with ML scores and calibrate thresholds using offline evaluation data.
Use batch + streaming updates and accept eventual consistency for real-time scoring.
Rules provide immediate coverage and clear explanations while ML models mature.
Keep feature reads local and precompute heavy features to avoid synchronous lookups.
Streams power real-time monitoring and offline model retraining without blocking scoring.
You're in the middle of an interview session. Leaving now will end your current attempt.
Explore concept overviews, real-system examples, key tradeoffs, and interview talking points for each roadmap section.
You've conquered this phase. These are the skills you now own:
