Graphsmedium45 minAI/ML Engineer

Eval Dependency DAG Recovery

AI/ML Engineer signal: topological sort + cycle detection in a LLM evaluation harness context. This is a ProdMatch-owned ai ml engineer drill, framed as a April 2026 Stripe AI Quality simulation, not a copied platform question.

Company context

Stripe · AI Quality

Freshness

April 2026

Product surface

LLM evaluation harness

ProdMatch interview simulation based on product-team patterns; not a claim of a real company question.

Question

You own LLM evaluation harness for a AI/ML Engineer loop. Given runs and directed dependencies between them, return a valid execution order. If a cycle exists, return an empty list and identify that the rollout is blocked.

Input

n nodes labelled 0..n-1 and dependency pairs [before, after].

Output

A valid order of all nodes, or [] when dependencies contain a cycle.

Constraints

1 <= n <= 200000
0 <= dependencies.length <= 400000
The graph may be disconnected.

Concepts

vector search
RAG retrieval
recommendation graphs
topological sort
cycle detection
dependency graph

runs: 4, deps: [[0,1],[0,2],[2,3]] -> [0,1,2,3] or [0,2,1,3]

Approach

Try framing your own approach first. The 30 seconds you think before peeking is where learning happens.

Clean Solution

Reveal the approach first.

How well did you understand?

Your rating tunes when this problem shows up again.

Common Mistakes

Mark visited when enqueueing, not after repeated dequeue.
Do not ignore disconnected components if the product surface can have them.

Next Similar Problems

Vector Dependency DAG Recoverymedium Vector Evidence Path Finderhard GraphRAG Dependency DAG Recoveryhard