RAG Interview Questions, issue 17, Jul 21, 2026
The Lexical Replacement Trap
Staff ML Engineer interview at OpenAI, and the interviewer asks:
“Your team wants to swap lexical query expansion for semantic expansion to boost precision. Walk me through why that’s the wrong way to do the migration.”
Don’t say: “Lexical expansion adds synonyms for recall, semantic expansion uses embeddings for intent, so we upgrade to the better one.”
Why upgrading your search stack to vectors isn't a fix, but a trade-off that silently drops high-intent queries, and the score fusion trick elite engineers use to survive it.
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.