Q41. When would you choose RAG over fine-tuning for a knowledge-base product? Walk me through the decision.
Type: Case | Difficulty: Medium
What they're testing: Whether you understand the actual tradeoffs or just parrot buzzwords. Real answer digs into cost, latency, and maintenance burden.
Framework hint:
Data freshness needs — RAG wins when knowledge updates weekly
Factuality requirements — RAG lets you cite sources
Cost per query — retrieval overhead vs training cost
Q48. Your product uses GPT-4 via API. What happens when OpenAI has an outage — how do you design for it?
Type: Case | Difficulty: Hard
What they're testing: Vendor risk mitigation for AI dependencies. Real senior AI PMs think about this before it happens.
Framework hint:
Fallback model ready to swap
Graceful degradation of AI-dependent flows
Feature flags
User-facing status communication
Cost tradeoff — hot vs cold standby
Q53. Your team wants to use a new AI startup's product because it's cutting-edge. What's your concern list?
Type: Case | Difficulty: Easy
What they're testing: Whether you push back on cutting-edge for cutting-edge's sake. AI startups fail fast.
Framework hint:
Funding + runway — 18 months?
Reversibility of swap
Data lock-in if shutdown
Enterprise readiness — SLAs, security, compliance
Roadmap dependency risk
Q58. Would you use an AI agent framework or write orchestration yourself?
Type: Case | Difficulty: Hard
What they're testing: 2025-current debate. Agent frameworks are maturing but often over-abstract.
Framework hint:
Orchestration complexity — simple chains don't need frameworks
Team AI/ML expertise
Debugging: custom vs framework internals
Iteration speed — swap components
Production reliability of many frameworks
🔓 16 more questions in this segment alone — including vendor evaluation design, multi-model routing, and healthcare-grade build vs buy calls. Get the full 100 on Gumroad →