← Back to experiences
[ OK ] Loaded —
[ INFO ]
$ cd
$ ls -lt
01
02
03
04
05
$ ls -lt
01
02
03
04
05
user@intervues:~/experiences/…$
Level: Senior-Level
Round: Full Journey · Type: Multiple Types · Difficulty: 6/10 · Duration: 360 min · Interviewer: Neutral
Topics: Machine Learning, Statistics, Algorithms, Coding, Behavioral Questions, Transformers, RNN, K-Means, Data Preprocessing, Feature Engineering, Model Evaluation, Optimization Algorithms
Location: Seattle, WA, US
Interview date: 2025-11-21
The first part focused on ML knowledge. The interviewer didn't ask basic definitions but instead delved deeply into my projects. Questions covered model selection, training details, and evaluation. Some typical questions were:
The interviewer also asked about the differences between Transformers and RNNs, their pros and cons with very long inputs, why attention can capture long-range dependencies, and why RNNs can't.
Another question was how to determine if two sets of images come from the same distribution.
The second part was ML coding. I had to write K-means by hand, without requiring it to be fully runnable. I just needed to explain the steps and ensure the matrix dimensions were correct. The interviewer mentioned that they don't test LeetCode problems, but others might.
Round 1: Statistics Application
Given two sets of user data (North America and Europe), each with multi-dimensional continuous features (e.g., dwell time, click-through rate, purchase conversion rate), how would I determine if the two groups have significant differences in their overall distribution?
Round 2: ML Deep Dive
This round continued to delve deeply into the projects I had worked on. The interviewer started with data preprocessing, asking why I chose that data cleaning method and feature engineering approach.
Then they continued to dig into evaluation, asking why I chose those metrics, what they reflected, how the validation set was divided, and whether I had considered stratified sampling.
Next, they asked which other models I would choose and had me analyze their pros and cons in terms of training time, inference latency, interpretability, and robustness to noise.
Finally, the interviewer asked from an optimization perspective about the pros and cons, and what the impact would be if I switched to Adam or RMSProp. The interviewer followed up by asking why not AdamW and in which scenarios it is more suitable.
Round 3: Coding
I was asked to do:
Round 4: HM Chat
This was a chat with the Hiring Manager, covering ML trivia.
Round 5
Behavioral questions.
LeetCode similar: LeetCode 56, LeetCode 347