Walk me through a model you trained that shipped. What was your eval strategy?

Strong candidates demonstrate: Has trained models that ship to production, Knows current frontier research (transformers, MoE, RLHF/DPO). Watch for these red flags: All Kaggle, no production, Outdated knowledge (mostly pre-LLM era).

How do you decide what to fine-tune vs use frontier APIs?

Strong candidates demonstrate: Has trained models that ship to production, Knows current frontier research (transformers, MoE, RLHF/DPO). Watch for these red flags: All Kaggle, no production, Outdated knowledge (mostly pre-LLM era).

How do you handle data drift?

Strong candidates demonstrate: Has trained models that ship to production, Knows current frontier research (transformers, MoE, RLHF/DPO). Watch for these red flags: All Kaggle, no production, Outdated knowledge (mostly pre-LLM era).

How would you serve a 70B model at $0.50 per million tokens?

Strong candidates demonstrate: Has trained models that ship to production, Knows current frontier research (transformers, MoE, RLHF/DPO). Watch for these red flags: All Kaggle, no production, Outdated knowledge (mostly pre-LLM era).

What's your favorite recent paper?

Strong candidates demonstrate: Has trained models that ship to production, Knows current frontier research (transformers, MoE, RLHF/DPO). Watch for these red flags: All Kaggle, no production, Outdated knowledge (mostly pre-LLM era).

Hiring guide · AI

How to hire a
ML engineer.

ML engineers (vs AI engineers) work closer to model training and infrastructure. Hiring well requires evaluating both research literacy and production engineering chops.

Comp range$200k-$400k base. Very senior frontier-lab people are higher.

Timeline12-20 weeks.

When to hire

Hire in-house

You're training your own models or doing serious data engineering for ML.

When to outsource

Outsource to a studio

You're using off-the-shelf APIs only. An AI engineer (not ML engineer) might be the right hire.

Evaluation · 02

Signs of strong candidates.

Strong signals

Has trained models that ship to production
Knows current frontier research (transformers, MoE, RLHF/DPO)
Production engineering chops alongside ML
Comfortable with PyTorch / JAX
Has dealt with data drift, evals, and A/B

Red flags

✗All Kaggle, no production
✗Outdated knowledge (mostly pre-LLM era)
✗No production engineering experience
✗Can't debug in PyTorch
✗No interest in ops or monitoring

Interview · 03

Questions to ask.

Q1
Walk me through a model you trained that shipped. What was your eval strategy?
Q2
How do you decide what to fine-tune vs use frontier APIs?
Q3
How do you handle data drift?
Q4
How would you serve a 70B model at $0.50 per million tokens?
Q5
What's your favorite recent paper?

Considering a studio?

We bridge
to your hire.

We help structure ML hiring briefs. If you're not ready to hire a ML engineer yet, brief us — we can fill the gap and help you hire later.

Brief us

Other hiring guides

How to hire a AI engineerAI

Need a ML engineer now?

Brief Vedwix. We can engage as a studio or help you hire one.

Talk to us

How to hire aML engineer.

Hire in-house

Outsource to a studio

Signs of strong candidates.

Strong signals

Red flags

Questions to ask.

We bridgeto your hire.

Need a ML engineer now?

How to hire a
ML engineer.

We bridge
to your hire.