Went through the PayPal MLE loop about two months ago, for a senior ML engineer role on the fraud detection team. Here's the real breakdown, not the sanitized version.
Rounds: recruiter call, ML coding screen, then a full virtual onsite with 5 rounds.
ML coding screen: they gave me a dataset (synthetic, about transaction history) and asked me to build a simple fraud classifier in Python. Sklearn, basic pipeline: preprocessing, feature engineering, model selection, evaluation. They care about your reasoning on feature choices and metric selection. Why F1 and not accuracy for fraud? Walk them through that. They're not expecting a gradient boosted tree with 94% AUC in 45 minutes. They want to see systematic thinking.
Onsite breakdown: Coding: data structures/algorithms, medium LeetCode level. One graph problem (BFS on a transaction network), one array manipulation. Nothing fancy. Know Python clean. ML system design: this was the meaty round. Design a real-time fraud detection system. Full scope: feature engineering from payment event streams, model training pipeline, real-time inference latency constraints (they said <100ms), model monitoring for concept drift. I talked about feature stores, online vs. offline feature computation, shadow deployment for new models. They went deep on the monitoring piece, how do you detect when the fraud pattern shifts after a regulatory change. Applied ML: they gave a scenario where your fraud model's precision drops over time. Walk through root cause analysis. This is a model debugging round basically. Know your confusion matrix mechanics and what precision/recall shifts signal. Behavioral: standard STAR stuff. One question specifically about a time you had to explain a model decision to a non-technical stakeholder who didn't trust it. Very applicable at PayPal where compliance and business teams are in the loop. Cross-team/leadership: they wanted to know how I'd work with data engineers to build feature pipelines and with product to define model success metrics. Less technical, more collaborative.
Offer: around $245k TC senior level, Bay Area. Stock was reasonable, not FAANG-tier but real.
If you're applying for the fraud/risk ML roles specifically, spend serious time on imbalanced classification, precision/recall trade-offs, and operational ML (model monitoring, retraining triggers). That's the whole game there.