FDEInterviews logo
RAG & Agent System Design / 56
hardNewGleanDatabricksOpenAI

The customer's corpus grew from one million documents to ten million and retrieval p99 tripled. Your code did not change. Find out why.

Everyone reaches for a bigger machine. The two mechanisms that actually cause this both scale with corpus size, both hit the tail before the median, and one of them is the permission filter your security review insisted on.

Updated Aug 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.

Everyone reaches for a bigger machine. The two mechanisms that actually cause this both scale with corpus size, both hit the tail before the median, and one of them is the permission filter your security review insisted on.

20 answers per topic instead of 10, plus saved progress and bookmarks · no cardor unlock all 506 remaining answers · ₹2,000 / $25
UP NEXT ON YOUR JOURNEY
FEDITOR'S NOTE

Strong answers open by noticing that p99 moved and p50 barely did. Candidates who treat that as noise miss the whole diagnosis, because both real causes here are distributional: they lengthen the worst queries first and leave the typical one alone. The second-order point worth listening for is that the ACL filter, which someone added for good reasons, gets more expensive every time the corpus grows while the user's own slice stays the same size, and it does so with no code change and no alert.

DISCUSSION · 0

No comments yet — be the first to share your approach.