Uber Software Engineer Interview Experience — Solved the Dependency Graph Problem, Still Rejected for Not Flagging Edge Cases

I finished the problem and ran a few test cases I wrote myself. A week later I got a "moving forward with other candidates" rejection.

I had AI polish up the problem statement a bit:

Package Build Order (dependency build order)

Problem description (abstract version):

The interviewer gave me an interface:

get_dependencies(package)

— it returns the direct dependencies of a given package.

Requirements: given a target package, output a valid build order such that all of its dependencies are built before it.

Constraints:

• The dependency relationship is a directed graph (A depends on B)
• The graph isn't given all at once — it's queried dynamically through the API
• You only need to consider the dependency subgraph relevant to the target
• If there's a cycle, return empty / raise an error
• The solution doesn't need to be unique

What this problem really is: this is a variant of a classic problem — topological sort on an implicit / lazily-loaded graph.

Added (2026-05-04 03:54 +08:00): the recruiter said the reason I didn't pass was that I wasn't proactive enough — I didn't think about what happens if the system goes down, and I didn't call out edge cases and failure cases.