25 AI engineer interview problems that cover 95% of the concepts, from RAG to agents
kabokablemolefe · x · 2026-10-13
A widely shared prep list of 25 problems claims to cover 95% of AI engineer interview concepts; an engineer in South Africa confirmed questions 1-5, 10-19, 23, and 24 came up in real local interviews.
Topics include:
- RAG: design for 10M documents, debug wrong answers, chunking for legal contracts, BM25 vs dense vs hybrid search, adding and evaluating a reranker
- Cost & performance: cut LLM cost 70% without hurting quality, reduce first-token latency, semantic cache with hit-rate tracking, routing between cheap and smart models
- Agents: reliable tool use, fixing stuck loops, memory design for long-running assistants