went through stripe's ML engineer interview loop in early 2026 for an L5-ish role on their fraud/risk ML team. there's basically nothing written about this so here's what i remember.
stripe's ML team is smaller than you'd think. most of the ML work is in payments fraud, risk scoring, and some internal tooling. they're not trying to build foundation models. if you want pure research or CV/NLP research roles, this isn't that.
the loop: recruiter screen (30 min): standard, what do you want, location, timeline, compensation ballpark technical phone screen (60 min): split between ML concepts and coding. the ML portion was about model evaluation: how do you evaluate a fraud classifier when the classes are this imbalanced? what metrics matter and which ones mislead you? the coding portion was a medium leetcode problem, nothing ML-specific onsite (4 panels): ml depth: deep dive on a project. not just "what did you do" but "why that model architecture", "what did you try that didn't work", "how did you measure production impact". they want you to have done real work, not just tuned hyperparameters ml system design: i got "design an ML system to detect fraudulent transactions at stripe's scale." lots of signal in this one. feature engineering, online vs. offline inference, latency requirements, handling concept drift, A/B testing the model safely in prod software engineering: one coding round, more standard data structures. they confirmed MLE at stripe is expected to write real production code, not just jupyter notebooks behavioral: tell me about a model you deployed that had unexpected behavior in production. what did you do. this was the best behavioral question i've gotten anywhere. it gets at whether you've actually shipped things.
what matters: imbalanced classification, fraud ML domain knowledge, model monitoring/observability. if you've worked in fraud or risk before this is a big plus being able to articulate tradeoffs between precision and recall in a business context, not just academically software eng is real. write clean code.
did get the offer. declined because of a better equity package elsewhere. comp was around base $215k, equity $500k/4yr for l5 in sf.