FDEBench Join us live

Q
5 / 159

General design

How would you design Netflix's search bar to return results for the query 'that show with the guy from Breaking Bad' without any exact keyword matches?

question bank FDEBench bank Source

01
Ask

What a strong candidate asks first

Before designing anything. Skipping this is the most common way the round is lost.

  • How common are queries like this — is this a tail case or a real segment?
  • Do we have structured catalog metadata (cast, crew), or only descriptions?
  • Is latency budget the same as normal search?

02
Rubric

What earns the points

6 criteria, 100 points total. Scored independently, partial credit allowed.

Sees it's a person, not keywords 20 pts

Identifies that the query references an entity indirectly, and that answering requires resolving the entity and traversing a relationship — not matching text.

If you miss it You jumped to a technique without diagnosing the query. This isn't a matching problem — the user named a person indirectly and wants things connected to them. Say that before choosing a tool.
Pins down which actor is meant 20 pts

Extracts the referenced entity from the query and resolves it to a canonical ID in the catalog, handling ambiguity.

If you miss it You never resolved "the guy from Breaking Bad" to an actual person in your catalog. That resolution step is the whole question.
Follows cast and crew links 20 pts

Uses structured relationships (cast/crew/franchise links) to find candidates, rather than hoping embeddings encode them.

If you miss it You relied on semantic similarity to encode "acted in." Embeddings capture topical similarity, not factual relationships — that needs structured data.
Knows why vector search fails 15 pts

Explicitly says why pure vector search fails here — the query and the target share no semantic content, only a factual link.

If you miss it You didn't say why vector search alone fails here. Being able to name the limits of the default tool is most of the signal in this question.
Ranks the actor's shows sensibly 15 pts

Ranks the resulting candidates on something meaningful — role prominence, popularity, user history — not arbitrary order.

If you miss it You found candidates but didn't rank them. An actor has many credits; which one the user means is a ranking problem.
Falls back to regular search 10 pts

Falls back to lexical or semantic search when entity resolution fails, rather than returning nothing.

If you miss it Your design returns nothing when entity resolution fails. Fall back to ordinary search — a mediocre result beats an empty page.

03
Flags

Red flags

Reported separately, not subtracted — this is the "what would have ended the interview" note.

  • Answers "use embeddings" and stops.
  • Assumes an LLM will just know the answer from training data.

—