System Design Interview

A repeatable framework for open-ended design problems. The goal is to show you can take an ambiguous prompt, narrow scope, propose options, and defend trade-offs — not to recite a “correct” diagram.

The Loop (≈45–60 min)

flowchart LR
  R[1. Requirements] --> E[2. Estimates]
  E --> A[3. API + Data model]
  A --> H[4. High-level design]
  H --> D[5. Deep dive]
  D --> B[6. Bottlenecks & trade-offs]

1. Define the problem (don’t skip)

• Ask questions to shrink scope: goals, who the clients are, what’s in/out of scope, existing systems to integrate with.
• State functional requirements, then non-functional ones (latency, availability, consistency, scale).

2. Capacity estimates (back-of-envelope)

• DAU → QPS (peak ≈ 2–10× average), read:write ratio, storage/year, bandwidth.
• These numbers justify later choices (cache? shard? CDN?).

3. API & data model

• Translate requirements into a few endpoints (send_message, get_messages).
• Pick the API style: REST, GraphQL, gRPC. Sketch the core entities and access patterns first (they drive the DB choice).

4. High-level design

• Layers: client → CDN/edge → API/LB → services → data stores → async workers/streaming.
• Place caches and queues deliberately; show the happy path end-to-end.

5. Deep dive

• Pick the 1–2 hardest parts and go deep: sharding strategy, hot keys, consistency model, idempotency, fan-out on read vs write.

6. Bottlenecks & trade-offs

• Single points of failure, the next scaling wall, and how you’d evolve. Name the trade-offs explicitly — that’s the signal interviewers want.

Toolbox to Reach For

• Scale reads: Caching, read replicas, CDN.
• Scale writes: sharding/partitioning, async via queues (Messaging & Event-Driven Architecture).
• Decouple: queues, pub/sub, event-driven.
• Consistency: know your CAP Theorem & Consistency position per data set.

Cross-Cutting (enterprise)

i18n · accessibility · security · caching · DB indexing · observability.

DORA-style KPIs

Deployment frequency · lead time for change · change-failure rate · mean time to recovery.

Questions Worth Asking

• How much of the role is building vs. designing/presenting?
• Do we have paying customers? What’s the real scale today vs. aspirational?