The ML Interview Loop, Round by Round

The ML Interview Loop, Round by Round

An ML engineering loop is not one interview repeated five times. Each round grades something different, and the mistake that costs the most offers is preparing for the coding round and treating the other four as conversation.

The recruiter screen

This round is a filter, and it is looking for two things: that your experience matches the posting closely enough to justify the team's time, and that you can describe it without rambling. Have a ninety-second version of your last two years ready. Most people have a five-minute version and get cut before the interesting part.

The coding round

For ML roles this splits in two. Some companies run a standard algorithms interview. Others ask you to implement k-means, or attention, or beam search from memory. Prepare for both. The second kind is far less commonly practiced and far more revealing.

The ML breadth round

Rapid questions across the whole field with no warning about which area. What is being graded is calibration: whether you can give a two-sentence answer when that is what was asked, and go three levels deeper when the follow-up wants it. Candidates who monologue on every question score badly even when every answer is correct.

The system design round

The round that decides your level. Forty-five minutes, a vague prompt, and a whiteboard. The structure matters more than any individual choice:

  • Clarify requirements and constraints in five minutes, not fifteen
  • Define online and offline metrics before proposing anything
  • Size the problem: traffic, latency budget, storage, cost
  • Sketch the architecture
  • State the tradeoffs before you are asked

Running out of time before the tradeoffs is the single most common way this round is lost.

The behavioral round

It is graded, and vague answers cost offers. Build a story bank with real numbers in it and stress-test each story against three follow-ups, because that is what a trained interviewer will do.

What to do with this

Audit yourself round by round rather than in aggregate. Almost everyone has one round that is quietly weak, and it is almost never the one they are practicing.

Keep reading

Your RAG System Has a Retrieval Problem
Your RAG System Has a Retrieval Problem

When the answer is wrong, teams reach for a better model. The document usually was never retrieved in the first place.

Read the Loss Curve: A Field Guide to Failed Runs
Read the Loss Curve: A Field Guide to Failed Runs

Loss flat, loss spiking, training down and validation up. Each shape has a short list of causes, and guessing is not one of the steps.

Gradient Boosting Is Still the Answer for Tabular Data
Gradient Boosting Is Still the Answer for Tabular Data

Reaching for a neural network on tabular data is a common way to look junior in a system design interview.