RAG Interview Questions, issue 21, Jul 25, 2026
The GraphRAG Scaling Trap
Senior ML Engineer interview at Google, and the interviewer asks:
“Your GraphRAG pipeline uses an LLM to extract entity-relationship triplets. It’s flawless in the demo. Now you’re ingesting 50K docs a week and your indexing bill is on fire. What did you trade away when you picked LLM extraction?”
Don’t say: “LLMs are more accurate than rule-based parsers, so it’s worth the cost.”
The hidden trap where appending documents to a knowledge graph destroys your deduplication, and why senior engineers trade pure LLMs for a hybrid, tiered pipeline.
The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.