writing2026-09-12

MOLOCO ML Engineer Intern: Screening Interview

writingblogcareerinterviewmoloco

Summary

Original article title: "MOLOCO ML Engineer Intern: Screening Interview." Only the filename's unsupported colon was replaced. Published 2026-03-23 · slug moloco-ml-screening-interview-2026 · category "Hiring, interviews, and careers / Recruitment processes and interview retrospectives" · source article recorded in Korean and English.
A retrospective on the screening interview for a three-month Machine Learning Engineer internship after the CodeSignal assessment. The roughly one-hour video interview focused on background and project explanations, followed by ML and CS fundamentals. Although the conversation seemed to go reasonably well, I received a rejection. The stated evaluation areas were ML knowledge, CS fundamentals, and fit for ML Modeling versus ML Infrastructure.
Nontechnical topics: interview preparation, a 30-second introduction, motivation for Moloco (advertising data and real-time systems with ML at their core), attitude toward rejection, and understanding of the role. After feedback that my job description was difficult to grasp at once, I was asked how I studied job postings and technical blogs. We also discussed office preference (Seoul first) and my question about relocating to US headquarters after a full-time conversion.
The longest discussion concerned Hanwha UGV communication anomaly detection: goals and constraints, including almost no labels; why Isolation Forest; the progression from rules to domain feature vectors and a cold-start rules/ML ensemble; contamination and isolated-network constraints. The interviewer probed what I meant by "built a model," distinguishing the complete system from the ML component. Other topics included undergraduate CS and operating systems (processes, threads, synchronization, deadlocks); LG Aimers boosting, XGBoost versus CatBoost, SMOTE-ENN, and imbalance metrics; QLoRA and 4-bit NF4 in Google ML Bootcamp; an advertising-conversion prediction scenario and PyTorch backward(); Modeling versus Infrastructure; and what could be done in a three-month internship.

Mentioned

Moloco: the prospective employer, styled MOLOCO in the blog.
Hanwha Aerospace internship (UGV-MON) and UGV-MON: the subject of the deepest questions.
LG Aimers 5th cohort: product defect detection and SMOTE-ENN questions. LG Aimers 7th cohort: the PatchTST and N-BEATS ensemble for food-and-beverage demand forecasting.
Google ML Bootcamp 2024 and financial_advisor: the Gemma QLoRA and 4-bit fine-tuning experience.
Previous article: MOLOCO ML Engineer Intern Coding Test.

Connections

Writing · Applications · Moloco ML Engineer Intern 2026
Atomic notes: Using a library and explaining its internals are different skills · Make a job description understandable at once, drawing on job postings and technical blogs · Define "built a model" precisely, distinguishing the system from the ML component · Feeling that an interview went well does not determine its outcome · Severe imbalance can benefit from combined sampling; SMOTE-ENN improved precision by eight percentage points over SMOTE alone · datetime.now()-based time series and mixed message types contaminate ML data; acknowledge prototype limitations
Concepts: 4-bit quantization (NF4) · CatBoost · Gemma-2B · Isolation Forest · N-BEATS · PatchTST · QLoRA · SMOTE-ENN · UGV communication protocols · XGBoost · i18n and multilingual support · Advertising metrics (CPI, CTR, ROAS) · Running a technical blog · Presentations · Demand forecasting · Time-series anomaly detection · Ensembles · Anomaly detection · Cold starts · Class imbalance · Supervised fine-tuning (SFT) · Isolated and offline environments · Feature engineering · Hyperparameter tuning

Local graph

Notes 2 · Links 1

Backlinks

MOLOCO ML Engineer Intern Coding Testwriting

Source posts

MOLOCO ML Engineer Intern Screening Interview: It Felt Like It Went Well, and I Was Still Rejected