Machine Learning Engineer · Full Journey · Multiple Types

IBM — Machine Learning Engineer ❌ Failed

Level: Unknown Level

Round: Full Journey · Type: Multiple Types · Difficulty: 6/10 · Duration: 180 min · Interviewer: Unfriendly

Topics: LLM, Graph Neural Networks, Visual Transformers, SQL, Pandas, Feature Engineering, Classification, System Design, Encoder-Transformers, BERT, T5, Llama

Location: Toronto, ON, CA

Interview date: 2025-04-06

Got offer: False

Summary

Round 1: Recruiter Screen

Question: Introduction of my resume. The interviewer was a hiring manager from France. We briefly chatted in French. I explained the biochemical background of my work with LLaMa and BERT, which he found interesting.

Round 2: Technical Round

Question: SQL queries, feature engineering with Pandas, classification with scikit-learn, and system design for a code assistant. I was a bit rusty with SQL and Pandas.

Round 3: Final Interview

Question: A barrage of rapid-fire questions about my resume, including technical questions. There were no behavioral questions and no opportunity for me to ask questions. The interviewer was a VP from India.

Details

I had a full interview process for a Data Scientist/Machine Learning Engineer role at IBM. I initially got the OA through a referral. The OA consisted of two deeply disguised LeetCode easy/medium problems related to DFS and binary search.

My first interview was with a hiring manager from France. We briefly chatted in French. The rest of the hour was spent with me introducing my resume. He was very interested in my experience with LLaMa and BERT, and I had to explain a lot of the underlying biochemistry due to my background being different from his.

I was then referred to another position and skipped the initial round and OA. This second round was with an interviewer from Quebec. The interview was 90 minutes long. The first part involved writing SQL queries. The second part was using Pandas for feature engineering and scikit-learn for classification to predict the likelihood of students enrolling in training programs based on their economic status and grades. The third part was a system design question for a code assistant. Although it had only been eight months since I left my previous job, I was a bit rusty with SQL and Pandas commands and occasionally had to consult documentation.

I was then invited to a final interview with two VPs. Only one VP, from India, attended. This VP asked me a barrage of questions, about 10 in total, in addition to 3 technical questions, without giving me a chance to fully answer each question or ask any questions myself.

I was ultimately rejected due to a "minor skill misalignment." I believe this was because I did not steer the conversation toward my LLM experience and instead focused on Graph Neural Networks and Visual Transformers, which were earlier in my resume. The interviewer asked me about recent encoder-transformer experience, and I only mentioned that I last deployed such a model in 2021 using T5, without mentioning my more recent experiences with GPT, BERT, and LLaMa.