Interview Gauntlet

10 questions. Every senior AI engineer is asked these.

You've built the knowledge. Now prove it under pressure. These are the exact questions asked at FAANG, top-tier AI companies, and forward-thinking enterprises when hiring Senior AI Engineers. No partial credit. Clock is running.
— Level 10 · Production RAG Pipeline

+200 XP8 min10 / 10

Chapter 1 Complete: Your RAG Toolkit

You've built and optimized a production RAG pipeline from first principles. Here's what you now know at a senior engineer level:

| Concept | Interview-Ready Answer | |---------|----------------------| | RAG vs Fine-tuning | RAG for dynamic knowledge + citations. Fine-tuning for style + classification. | | Chunking | RecursiveCharacterTextSplitter at 512 tokens, 50 overlap, always parse HTML/PDF first. | | Embeddings | Self-host BGE-M3 at scale. text-embedding-3-small for startups. Matryoshka at 512 dims. | | Vector DB | pgvector for <100M vectors if you run Postgres. Pinecone for serverless billion-scale. | | Hybrid Search | Dense + BM25 + RRF. Pushes precision from 62% to 84%. Zero extra infra on Postgres. | | Re-ranking | Conditional skip at confidence > 0.85. Saves 63% of cross-encoder calls. | | Evaluation | RAGAS faithfulness >= 0.80 as hard gate. Cross-family judge. Run on every PR. | | Cost | Semantic cache (#1 lever), model routing (#2), Matryoshka dims (#3). |

These 8 concepts cover 90% of Senior AI Engineer RAG interview questions.

1 of 4