Chime · Primly Community

Chime machine learning engineer interview: what the loop actually looks like in 2026

ds_dmitri · 5 replies

went through the chime MLE loop this spring, L4/L5 level. sharing because i found almost nothing useful when i prepped.

the process was: recruiter call, then a 45-min technical screen, then a 4-round virtual onsite. total clock was about 5 weeks, which felt long for a fintech company but wasn't the worst i've seen.

phone screen: one coding problem (medium, array manipulation) and about 15 minutes of ML fundamentals. the interviewer asked me to explain precision vs recall in the context of fraud detection, which made sense given their core product. nothing trick-y, but you have to actually know what you're talking about.

onsite: round 1: ML system design. they gave me a prompt around building a transaction risk scoring system. this is where chime is actually interesting to interview with because the domain is real. they care about latency (real-time scoring at checkout), data skew (fraud is rare), and model explainability for regulatory reasons. i spent a lot of time talking about feature engineering from transaction sequences, using something like gradient boosting vs a small neural net for interpretability, and how you'd version and A/B test the model without leaking labels. they seemed genuinely engaged. round 2: coding. two problems, both medium difficulty on leetcode scale. one was sliding window, one was graph-adjacent (BFS variant). 45 min, pretty standard. round 3: ML depth. they picked something from my resume (in my case, a recommender system i'd built) and drilled down for 40 minutes. be prepared for real follow-up questions, not just surface stuff. round 4: behavioral. roughly 5 questions. the usual STAR format. they asked about a time i disagreed with a stakeholder on a modeling approach, and about handling ambiguous problem definitions. standard PM-adjacent questions.

verdict: the debrief took about 2 weeks. i got an offer. the loop was reasonably well-run. the ML system design round is the hardest part and is where i think most candidates get tripped up if they haven't thought about fintech-specific constraints (latency, compliance, fraud label lag).

feel free to ask questions.

5 replies

ml_mike

that fraud scoring prompt is almost word-for-word what i got at a different fintech company last year. it's becoming a standard design question for the space. the latency angle is the key differentiator. most candidates design for batch, chime's actual scoring is meant to be synchronous at checkout so you have maybe 300ms total. if you pitch a pipeline that calls a heavy neural net you're going to get pushback.

ds_dmitri

yep, exactly. i talked about keeping a lightweight GBDT online-scoring service and doing heavier retraining async. the interviewer seemed to know the tradeoffs well, it felt like a real conversation not just a gotcha.

newgrad_neil

how strictly are they hiring to L4 vs L5? i'm a new grad with an ML focus but interviewing for a role titled 'ML Engineer' without a level specified. did they tell you the level going in or was it calibrated during the loop?

visa_vik

do they sponsor H1B? that's my main gating question before spending 5 weeks on a loop.

tired_recruiter

chime does sponsor but their process has gotten more deliberate about it since 2024. make sure you ask the recruiter explicitly on the first call, don't assume based on the job posting.