Went through the KPMG ML engineer interview process about six weeks ago. They're expanding their data and AI practice fast, especially on the consulting side where clients want AI-enabled products built quickly. Here's what the loop looked like.
Background on the role: It's within their Lighthouse AI / data advisory practice. Most of the work is client-facing: building ML systems for financial services, healthcare, or public sector clients. Not research. Not pushing state of the art. Applied ML with real deployment constraints.
The rounds:
Phone screen with recruiter. Standard. Wanted to understand my background and why consulting vs. product. Have a coherent answer for the consulting question.
Technical screening call with a data scientist. 45 minutes. This was mostly about my ML fundamentals: bias-variance tradeoff, regularization, how I'd explain a gradient boosting model to a client who has a compliance team, feature engineering for tabular financial data. No code in this round. Mostly conversation.
Live technical round. 90 minutes. Two parts. First half: a case-style scenario. Client has a classification problem, fraud detection, imbalanced labels, not a huge dataset. Walk through your approach end to end. I talked through EDA, baseline models, handling class imbalance, evaluation metrics and why accuracy is the wrong metric here. They pushed back well, it wasn't a gotcha interview, more like a working session. Second half: they gave me a Jupyter notebook with some messy data and asked me to clean it and build a simple logistic regression baseline. Python expected. Pandas, scikit-learn. They weren't evaluating my speed, more my habits: did I check for leakage, did I log my experiments, did I comment what I was doing and why.
Behavioral round. 60 min. All STAR. Consulting-framed. Best question: describe a time you had to convince a skeptical client stakeholder to adopt a model you built. They want to see you can translate ML outputs to business decisions.
Comp: My offer was around $130k base at a mid-senior level, in Chicago. Bonus-eligible. Lower than big tech ML roles but consulting gives you client variety that's genuinely useful for building a portfolio of problem types early in a career.
Happy to answer follow-up questions if anyone else is prepping for this track.