HackTheRounds Interview Experiences

Stripe MLE OA 2026: Machine Learning Online Assessment Questions & Prep Guide

Stripe's 2026 MLE Online Assessment: CNN traffic sign classification in PyTorch and pandas employee attendance analysis.

By HackTheRounds Team · 2026-03-28

Overview

Stripe's 2026 MLE (Machine Learning Engineer) OA is different from typical SWE assessments. Instead of algorithm puzzles, you get practical ML implementation tasks — building models in PyTorch and doing data analysis with pandas.

Problem 1: Traffic Speed Limit Sign Classification (PyTorch)

Topics: CNN, Image Classification, Deep Learning

Build a neural network to classify speed limit signs into 3 categories (30km/h, 70km/h, 120km/h) using provided training and test datasets.

What they expect: - A working CNN model in PyTorch (not TensorFlow) - Proper data loading, augmentation, and train/val split - CSV output with image paths and predicted labels (0, 1, 2) - Evaluated on accuracy

Approach: Standard CNN pipeline — Conv2d layers with ReLU/BatchNorm, MaxPool, FC layers. Data augmentation (random rotation, flip) helps significantly. Even a simple 3-conv-layer network can hit 95%+ accuracy on this dataset.

Tip: Don't over-engineer. A ResNet-18 with transfer learning would work but isn't necessary. Focus on getting a clean, working pipeline.

Problem 2: Employee Attendance Analysis (Pandas)

Topics: Data Manipulation, Pandas, Aggregation

Filter employees who visited sales branches at least 3 times within a month. Key constraint: multiple visits on the same day count as 1.

Approach: 1. Parse dates and extract year-month 2. Drop duplicate (employee id, branch, date) rows 3. Group by (employee id, branch, year month) and count 4. Filter for count = 3 5. Return as DataFrame

Tip: The "same-day = 1 visit" constraint is the gotcha. Many candidates forget to deduplicate before counting.

Stripe MLE Prep Tips

  1. Know PyTorch well — they specifically test PyTorch, not just "any ML framework"
  2. Pandas fluency is mandatory — groupby, merge, pivot, and aggregation should be second nature
  3. Practical over theoretical — they want working models, not research-paper-quality architectures
  4. Code quality matters — even in an OA, Stripe evaluates how readable and well-organized your code is
  5. After OA: Expect coding + system design focused on ML infrastructure (feature stores, model serving, A/B testing)

Explore Stripe interview questions on HackTheRounds.