The governing idea is simple: production systems will fail; your job is to design for graceful failure, fast recovery, and continuous learning.

Three pillars: Reliability, Resilience, Performance.

• Pillar 1 — Reliability
• Pillar 2 — Resilience
• Pillar 3 — Performance

How to Use This in an Interview

When you sketch any architecture and the interviewer asks "how do you keep this reliable in production?", your answer should hit:

1. SLO for each critical path (latency, error rate)
2. Timeouts on every external call
3. Circuit breakers around your LLM and vector DB
4. Retry with exponential backoff + jitter
5. Graceful degradation path (cached fallback, simplified response)
6. Bulkheads if multi-tenant
7. Blameless postmortem process when things fail
8. Distributed tracing so you can diagnose fast

You don't need all eight in every answer — but touching 4–5 unprompted signals strong operational maturity.