Cisco · Primly Community

Cisco data scientist interview (SQL + case + stats), full breakdown of what I faced

ds_dmitri · 4 replies

Interviewed for a data scientist role at Cisco on the product analytics team earlier this year. Three rounds of technical content across the loop. Sharing specifics on the SQL, case, and stats components because the DS interview at enterprise companies varies a lot.

SQL round (45 min): done live on a shared SQL environment. Two questions. First was a join-heavy query involving network event logs: find users (devices, in this context) who triggered a specific alert sequence within a 30-minute window. Window functions, conditional aggregation, GROUP BY. Medium complexity but the nested logic took me a bit to map out cleanly. Second question was about data quality: given a table with expected and observed metric values, write a query to surface anomalies above a threshold. Straightforward once you read it carefully.

Stats / ML case (60 min): the prompt was roughly: you're analyzing whether a new network security feature reduces false positive alerts. How would you design the analysis? They walked me through this as a discussion, not a coding problem. I talked through A/B test design, why randomization is hard when your units are enterprise accounts (not individual users), how I'd handle network effects between accounts on shared infrastructure, and what my primary metric would be. They pushed back on my choice of metric twice, which I think was a good sign.

They also asked a standalone stats question: "explain the difference between precision and recall to a non-technical stakeholder and when you'd optimize for each." Not a gotcha but they wanted a real-world example, not a textbook definition.

Product sense / case (30 min): a brief version of a product analytics case. "AppDynamics usage has dropped 10% in the last quarter. How do you diagnose it?" Funnel decomposition, user segment breakdowns, feature-level attribution. Classic DS case structure.

What to prep: window functions and self-joins in SQL, experimental design basics for non-iid settings, and a solid example of a real model or analysis you shipped and what happened after. They don't ask deep ML modeling at this team; it's more analytics and experimentation.

I got an offer. Happy to answer follow-up questions.

4 replies

analyst_ana

The experimental design question for non-iid settings is exactly the gap I always feel nervous about. How deep did they go on the network effects problem? Like did they expect a specific solution or were they looking for awareness that it's a problem?

ds_dmitri

Mostly awareness plus a directional approach. I mentioned cluster-based randomization at the account level and they seemed satisfied. They weren't looking for a switchback design or a synthetic control deep dive. Knowing the problem exists and having a reasonable first-order solution was enough at this stage.

numbers_only

Congrats on the offer. Any color on comp for the DS role, level, location?

marketer_mei

The AppDynamics question is interesting. They're basically testing whether you know their product portfolio well enough to make a sensible diagnosis. Do you think it helped that you knew what AppDynamics actually does before going in?