Workday · Primly Community

Workday machine learning engineer interview: it's not what you'd expect

ml_mike · 6 replies

Went through the Workday MLE loop in Q1 2026. Wanted to write this up because the setup is pretty different from what people usually associate with MLE interviews at product companies.

Role: Machine Learning Engineer on their Skills Cloud and talent intelligence team. Not a core platform MLE. This matters because the interview skews toward applied ML, not research.

The loop Recruiter call. Fine. Technical screen (60 min): Two parts. First 30 minutes was a Python coding question (clean up and process a messy dataset, write a function to compute feature statistics). Second 30 minutes was ML concept discussion: explain precision vs. recall tradeoffs, when would you use a tree-based model vs. gradient boosting, walk me through how you'd evaluate a ranking model. Onsite (5 rounds): ML system design: design a job-skills matching system. They gave me a product context (Workday's talent marketplace) and asked me to design end to end. Feature engineering, model choice, training pipeline, serving, monitoring. Coding: LC medium, two questions. Graph search and a string manipulation problem. Felt more like SWE screen than ML. Applied ML deep dive: walk through a real model you've built. They asked detailed follow-up questions. I talked about a recsys project and they dug into my offline eval approach. ML theory: bias-variance tradeoff, regularization, embedding representations. Nothing exotic but requires more than surface knowledge. Behavioral: 4 questions, all STAR.

What they care about at Workday for MLE

Skills taxonomy and NLP are central to their AI products. If you have background in entity matching, embedding models for semantic similarity, or any kind of people/jobs/skills graph, mention it. I spent about 10 minutes on my word2vec and transformer work and it was clearly relevant.

They do NOT want pure research profiles. If your experience is mostly academic benchmarks and papers with minimal production work, it'll be a tough fit. They want someone who can take a model to a production API endpoint and keep it healthy.

Comp: $185k base (Pleasanton), equity grant TBD at offer, mid-senior IC level.

6 replies

consultant_cam

The job-skills matching system design is a classic Workday ML domain question. I'd prep by understanding how skills taxonomies work (ONET, ESCO, LinkedIn's skill graph), because the features you'd engineer from those are very different from general recsys work.

sec_sasha

Did the ML system design round include any questions about data privacy or model fairness? Given that this is HR data (people's skills, performance), I'd expect some sensitivity there.

ml_mike

Yes, they asked about bias in skills recommendations specifically. Like: if your model trained on historical hiring data, what could go wrong. I talked through demographic parity, disparate impact, and how I'd audit the model outputs. They clearly care about this given the regulatory environment around HR tech.

quietquit_quincy

The fact that they included LC medium SWE coding in an MLE loop is annoying but unsurprising. Enterprise companies treat MLE as more of a hybrid role than pure research shops do.

growth_gabe

How much did they get into ML infrastructure vs. modeling? Like did they ask about model serving, latency constraints, feature stores? Or was it all modeling focused?

ml_mike

Some, not deep. They wanted to know I understood the serving side (batch vs. real-time, latency tradeoffs) but didn't quiz me on specific tools. Knowing the concepts was enough.