← 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: Mid-Level
Round: Full Journey · Type: Multiple Types · Difficulty: 5/10 · Duration: 240 min · Interviewer: Unfriendly
Topics: Coding, Behavioral, Machine Learning Fundamentals, Applied Machine Learning, Distributed Computing, Data Structures, Algorithms, System Design
Location: San Francisco Bay Area
Interview date: 2026-02-05
Question: Find the top 10 most frequent IPs from a log file.
Question: Standard behavioral questions focusing on unique projects, working under deadlines, and improving team culture.
Question: Variation of Basic Calculator II with added rules and constraints.
Question: Questions on ML fundamentals such as Bias-Variance Tradeoff, Logistic Regression, Gradient Descent, Cross-Entropy, and Regularization. Discussion of projects.
I had previously interned in the same role in 2025 but didn't match with a team. This time, as a new grad, I reapplied and contacted the HR representative from before.
Round 1 Coding: The coding question involved finding the top 10 most frequent IPs from a log file. My approach was to use frequency counting and then apply a TopK algorithm. The interviewer asked follow-up questions such as:
Because I had prior experience with distributed computing and storage optimization projects, the interviewer was quite interested in those follow-up questions.
Round 2 Coding (Onsite): The problem was a variation of the Basic Calculator II problem. It had added rules and constraints, such as new empty case handling. The interviewer was very detail-oriented. I had to explain my approach clearly on the whiteboard before coding. I also had to confirm edge cases and clearly explain the complexity. There weren't many follow-up questions in this round. The focus was on explaining the main solution clearly, implementing it completely, and ensuring the logic was rigorous. The overall pace was steady, not focused on speed, but more on the thought process and explanation.
Behavioral Questions: The behavioral questions were fairly standard, but I felt like I spoke a lot and my mouth was dry. Questions I remember include: Most unique and rewarding project, working under strict deadlines, and how I would improve team culture as an individual contributor. The questions themselves weren't difficult, but the follow-up questions delved deeper into the project and the details of my answers.
Machine Learning Domain Questions: The questions were focused on machine learning fundamentals and applied understanding. The questions themselves weren't particularly difficult, but they emphasized whether I truly understood the underlying logic, rather than just reciting memorized answers. Topics included: Bias-Variance Tradeoff, Logistic Regression, Gradient Descent, Cross-Entropy, L1 / L2 Regularization, and Forward / Backward Propagation. For example, for L1/L2 regularization, the interviewer didn't just ask for the differences, but also asked why L1 produces sparsity and how to explain it from a geometric perspective, requiring me to draw constraint shapes to illustrate why 0 is more likely to occur. I also had to constantly confirm my understanding of the questions with the interviewer to avoid deviating from the intended direction. For Forward / Backward Propagation, I needed to explain the entire calculation process, loss definition, how gradients are passed through the chain rule, and the parameter update logic. I basically explained and wrote formulas and drew diagrams on the whiteboard while extracting the core logic on a document. The follow-up questions constantly pursued a deeper understanding.
I also discussed my previous internship and PhD projects.