JPMorgan Chase · Primly Community

JPMorgan Chase machine learning engineer interview, what they focus on in 2026

ml_mike · 4 replies

Did the JPMC ML engineer loop a few months ago for their AI and ML group within the CIB (corporate and investment banking) tech division. Writing this up because most ML interview content is for Google/Meta and the financial services angle is different.

First: be clear on whether you're interviewing for ML engineer (builds the systems) vs. data scientist (builds the models). The distinction matters at JPMC and the interviews are different. I was MLE track.

Phone screen with the hiring manager. 30 minutes, mostly background and fit. He was direct: they use ML primarily for anomaly detection, transaction classification, NLP on documents (think loan agreements, research reports), and increasingly for risk modeling. He wanted to know if I'd be bored. Honest answer: it depends what you find interesting. If you want cutting-edge generative AI work, this is probably not the right place. If you want reliable, production-grade ML systems at massive scale in a high-stakes domain, it's genuinely interesting.

Technical screen was 60 minutes. Two parts: coding and ML system design. The coding was a Python problem involving vectorized operations on a dataset (no NumPy allowed, had to implement from scratch). The ML system design was: design a fraud detection model that updates in near-real time. I covered feature stores, retraining cadence, model serving latency requirements, and drift detection. They liked that I talked about explainability as a requirement in a regulated environment.

ML depth round. This was the one I found most interesting. Deep questions on specific topics: class imbalance in fraud detection datasets (we talked about SMOTE vs. cost-sensitive learning vs. threshold tuning), feature engineering for time-series transaction data, how to evaluate a recall-sensitive model where false negatives are expensive. These were not surface-level. They wanted to see that you've actually trained models and dealt with the messy real-world stuff.

Coding round. LeetCode medium style, algorithm plus data structures. Two problems, 60 minutes.

Behavioral round. Standard STAR. Three questions. The one I remember: "Tell me about a time a model you deployed had unexpected behavior in production. What happened and what did you change?"

Offer for senior MLE in NYC was around $160-175k base with bonus. Smaller equity component than tech companies but the base is solid. They moved me through the loop in about five weeks total.

4 replies

ds_dmitri

The explainability point is important and underrated. Banks are regulated and "we can't explain why the model denied this loan" is not an acceptable answer for compliance or legal. If you're going into any ML interview at a financial institution, have a real answer on interpretability: SHAP, LIME, monotonic constraints, model cards.

market_realist

Honest question: is the work actually intellectually interesting once you're in? Fraud detection and document NLP sound fine but I've heard big bank ML teams move slowly due to infrastructure and approval processes.

ml_mike

Depends on the team. The CIB AI group I interviewed with seemed to have decent autonomy. They mentioned running on AWS with some internal tooling. But I've heard the risk and compliance approval cycle for putting a new model in production at JPMC can be long. That's the real tradeoff.

laidoff_lena

Thank you for calling out the generative AI point directly. I see too many candidates walk into bank ML interviews expecting to talk about LLM fine-tuning and then seem disappointed when the job is production reliability and feature engineering. Set expectations correctly.