Target · Primly Community

Target machine learning engineer interview, honest take from someone who almost didn't take the call

ds_dmitri · 5 replies

I'll be upfront: when Target's recruiter messaged me I nearly passed. I associated them with retail, not ML. That was wrong and I want to save other ML engineers from the same lazy assumption.

Target runs recommendation systems at real scale. Personalized search ranking, same-day delivery routing models, dynamic pricing signals, demand forecasting for 2000+ stores. The team I interviewed with was their Personalization Platform group.

My loop for an MLE (senior level equivalent):

Recruiter + HM screen (separate calls): HM screen was legit technical. He asked about my recsys experience, how I've approached cold-start problems, and whether I'd worked with real-time feature serving infrastructure. This wasn't a filter round disguised as a conversation, he actually wanted to know.

Onsite (4 rounds):

ML depth round: Two-hour session with two senior MLEs. They started with how I'd approach ranking product search results for a logged-out user (no interaction history, only query + catalog features). We went deep on feature engineering choices, the tradeoff between collaborative filtering and content-based approaches for cold users, and how I'd evaluate model quality offline before deployment. One of them asked how you detect when a model's online/offline metric correlation breaks down. Good question. Real operational concern.

Coding round: Python. One problem involving graph traversal (BFS on a store layout graph, honestly not what I expected from a recsys ML role, but fair enough). One pandas/numpy data manipulation problem. Clean code mattered more than trick optimization.

System design (ML system): Design a feature store to support real-time personalization. I went through ingestion, serving latency requirements, offline/online parity, feature versioning and rollback. They probed hard on the serving latency part, what's your SLA for a product page recommendation, how do you get there with feature retrieval.

Behavioral: Standard seniority-calibration questions. When did you push back on a project direction, when did you make a call under uncertainty without full data. They wanted concrete examples not philosophy.

Assessment: This is a real ML team working on real problems. They're not hiring people to tune hyperparameters on simple models. If you come in with solid recsys or ranking experience and can talk MLOps (feature stores, model monitoring, A/B testing infrastructure), you'll have a real conversation. Comp was lower than equivalent FAANG roles, but not by as much as I expected, and the problems are legitimately interesting.

5 replies

mobile_mara

The cold-start question is classic for any recsys interview but retail cold-start is particularly messy. You have logged-out users, incognito sessions, users with only one order two years ago. How did you frame your answer for logged-out specifically?

ml_mike

I framed it around three signal tiers: session context (query text, device, time of day, page sequence), catalog features (product attributes, category, price tier), and aggregate popularity with temporal decay. Then talked about a lightweight context encoder to embed session signals for ranking. The "no interaction history" constraint matters a lot for how you weight catalog vs popularity features. They seemed to like the structured decomposition.

qa_quinn

Any comp numbers? Senior MLE at Target, Minneapolis vs remote?

ml_mike

My offer was Minneapolis-based. Base in the $155-170k range (they have bands), bonus around 12-15% target, equity vest over 4 years. Total comp considerably below comparable FAANG/MANGA offers but the team seemed genuinely good. I didn't take it but it was a closer call than I expected going in.

sre_sol

The feature store system design question is increasingly common at any company doing real-time ML. If you're prepping for Target ML or any retailer with personalization, I'd get very comfortable with Feast, Tecton, or at least the conceptual architecture of online/offline stores.