AI Engineer Interview Scenarios

Scenario — Production RAG at scale

You have millions of documents and need a production RAG system with predictable latency.

Cover:

• Requirements and latency budget.
• Ingestion and document update strategy.
• Chunking, embeddings, retrieval, reranking, and context construction.
• Evaluation and grounding.
• Security and tenant isolation.
• Cost and failure modes.

Follow-up: What would you change if retrieval quality is poor but latency is already at the budget?

Scenario — Hallucinations

A model produces confident answers that are not supported by retrieved documents.

Explain how you would isolate retrieval quality, prompt/context construction, model behavior, evaluation quality, and fallback behavior.

Strong answers connect: requirements → evidence → architecture → trade-offs → verification.