Skip to content
Vera
Interview Prep

Defend your model in production, not in a notebook.

Training pipelines, drift, evaluation choices, and the ML system design conversation.

ML loops fail on the systems half, and the systems half is entirely verbal.

What Vera actually is

A phone call with someone who is not there. You say the thing out loud, they answer the way a real person would, and you get to do it again — as many times as it takes, with nobody listening.

1She calls youA real phone call, about ten seconds after you tap. Lock screen, headphones, car — wherever you take calls.
2She plays the other personNot a coach reading tips. Machine Learning Engineer Interview Prep, with someone on the other end who answers back.
3She pushes when you are vagueFollow-up questions, interruptions when you ramble, and a straight read at the end on what landed.
Practice ML Rounds

15 minutes free · no card needed

Practice ML Rounds

15 minutes free · no card needed

“What if I am bad at it?”Everyone is, on the first one. That is the entire reason it happens here and not there.
“What if I do not know what to say?”She asks first and follows up on whatever you answer. You never start from a blank page.
“What if it is awkward?”It is, for about twenty seconds. Then it is a conversation, and nobody heard the twenty seconds.

Not practising doesn’t save you anything.

The ticket, the evening, the weeks of applying — all of it is spent before anyone says a word. The practice is the only cheap part, and the only part that changes how the rest goes.

Practice ML Rounds

15 minutes free · no card needed

What she'll actually ask

She asks these the way a real person would, then follows up on whatever you say.

  1. Walk me through how you would serve this model at scale.
  2. Your offline metric improved and the online metric did not. What now?
  3. How do you detect and handle model drift?
  4. Why did you pick that evaluation metric over the obvious alternative?
  5. Tell me about a model you shipped that caused a problem downstream.
More about this one

Machine learning engineer interviews sit between research and infrastructure, and the rounds that trip people up are conversational: how you would serve this model, what breaks in production, why you chose that evaluation metric, what you do when the offline win does not reproduce online. Vera runs those rounds over the phone, pressing on the operational side that research-heavy candidates skip and the modeling side that infra-heavy candidates gloss.

What she listens for

Offline-online gapWhether you volunteer training-serving skew before being asked.
Evaluation reasoningPicking a metric without naming what it fails to capture.
Operational thinkingRetraining cadence, monitoring, and rollback rarely come up unprompted from candidates.
  • model drift
  • training-serving skew
  • feature store
  • offline evaluation
  • inference latency
  • retraining pipeline
  • ML system design
  • shadow deployment
“I could talk about architectures forever and had never once rehearsed 'how do you know it is still working next month.' That was half the interview.”
Ibrahim S., ML Engineer

Related practice

Data Scientist Interview PrepExperiment design, causal reasoning, and defending an analysis out loud.System Design Interview PracticePractice narrating a design, defending tradeoffs, and handling the follow-up that changes everything.Data Engineer Interview PrepData modeling, orchestration failures, and the tradeoffs behind your warehouse design.Software Engineer Interview PrepBehavioral rounds, project deep dives, and the "tell me about a technical decision" question — rehearsed until they sound natural.

Common questions

Can Vera do ML system design?

Yes, verbally — which is how the round runs anyway. You will not have a whiteboard, so it forces clean verbal structure.

Is this for research scientist roles?

Partly. Research loops lean on paper discussion and depth; this page targets applied ML and production-facing roles.

Should I practice explaining my own past models?

Yes, and it is the highest-value use. Tell Vera the project and let it interrogate your choices.

What if I am bad at it?

Everyone is, on the first one. That is the entire reason it happens here and not there.

What if I do not know what to say?

She asks the first question and follows up on whatever you answer. You never start from a blank page.

What if it is awkward?

It is, for about twenty seconds. Then it is a conversation, and nobody heard the twenty seconds.

Ready to try it?

Your first 15 minutes are free. No card required.

Practice ML Rounds