How to hire an AI/ML engineer who ships models, not just demos
To hire an AI/ML engineer, first decide whether you need research depth or production engineering, because the pools barely overlap. Source from public work like Hugging Face, Kaggle and open source, then test evaluation discipline. The Cognitive searches ~900M public profiles and runs a live AI video interview as the first round.
Which kind of AI/ML engineer are you hiring?
Most failed AI hires come from mixing these up. Pick 1 column before you write the job description or the search.
| Profile | Research ML | Applied ML | LLM application |
|---|---|---|---|
| Main output | New methods, papers, trained models | Models in production with a measured effect | Features built on hosted or open models |
| Typical background | PhD or research lab | Software or data engineering, then ML | Backend engineering, then LLMs |
| Public trail | Papers, conference talks, arXiv | Kaggle, open source, blog posts | GitHub repos, eval tooling, Hugging Face |
| Core skill to test | Experimental rigor | Serving, monitoring, drift | Retrieval, evals, cost per request |
| Common miss | Hiring a researcher to run production | Hiring for model knowledge, missing data work | Hiring prompt demos without evals |
Where do AI and ML engineers work, and how do you find them?
ML engineers leave more public evidence than most engineers. Hugging Face profiles show models and datasets they published. Kaggle shows competition results. GitHub shows contributions to serving and evaluation projects such as vLLM, MLflow or eval harnesses. Conference workshop papers and the MLOps Community show who writes about production problems. Any of these says more than a resume line reading "AI".
Titles are noisy. Since 2023, many backend engineers added "AI engineer" after one retrieval project, while strong applied ML people still call themselves data scientists or software engineers. Search on what they shipped and measured, then use the title only to break ties.
Search strings written for AI/ML engineers
These target the applied and LLM columns, which most companies need. Swap the tool names for your stack.
BOOLEAN
("machine learning engineer" OR "ML engineer" OR "AI engineer" OR "applied scientist" OR "data scientist") AND (PyTorch OR TensorFlow OR JAX) AND (production OR serving OR deployed OR inference) AND (MLflow OR "Weights & Biases" OR vLLM OR Triton OR SageMaker OR Kubernetes)
GOOGLE X-RAY
site:huggingface.co "fine-tuned" "evaluation" -course
site:github.com ("RAG" OR "retrieval") "eval" "latency" README
PLAIN-ENGLISH BRIEF
ML engineer with 4+ years, at least 2 putting models or LLM features into production. Must have built an evaluation set and monitored quality after launch. Has served models behind an API with a latency budget. Python and PyTorch. Hybrid in London.
Why speed matters more here
Strong applied ML engineers usually run several processes at once. A 3 week gap between first contact and first interview loses people who would have said yes.
What does the AI/ML engineer market look like in 2026?
BLS has no ML engineer code. These are the 3 nearest occupations from the Occupational Outlook Handbook, read 5 October 2026.
| Occupation (SOC) | Median wage, May 2025 | Projected change, 2025 to 2035 | Jobs, 2025 |
|---|---|---|---|
| Data scientists (15-2051) | $120,230 | 35% | 275,600 |
| Computer and information research scientists (15-1221) | $140,300 | 22% | 38,600 |
| Software developers (15-1252) | $135,980 | 10% (with QA and testers) | 1,905,400 (with QA and testers) |
Source: BLS Occupational Outlook Handbook pages for each occupation, last modified 27 August 2026. Research scientists typically need a master's degree.
What the BLS figures mean for an ML hire
The research scientist code is small, about 38,600 jobs, and BLS lists a master's degree as typical entry. If your role needs that profile, expect a thin pool. Data scientists are 7 times as numerous and growing fastest of the 3, which is where many applied ML engineers start. Our reading: for most product companies, the practical pool is applied and LLM engineers from data science and backend engineering.
What should you ask an AI/ML engineer in an interview?
Five competencies for applied and LLM roles, with a question each. Evaluation runs through every one of them.
Tell me about a model or LLM feature you put into production. How did you know it was good enough to ship?
An eval set, a threshold agreed before launch, and monitoring afterwards. Production engineers say this without being asked twice.
What was in the eval set, and what did you do when quality dropped?
How did you put a model behind an API?
Request batching, a GPU or CPU decision with a reason, a latency budget, and a stack such as Triton, vLLM or SageMaker.
Walk me through a retrieval system you built.
How they chunked documents, why they picked an embedding model, and how an eval set measured answer quality rather than a few demo prompts.
What failure did the eval set catch that demos missed?
How did you build and split your training data?
Labeling guidelines, deduplication, and splits by user or time so test data never leaked into training.
When did you choose fine-tuning over prompting, or a smaller model over a larger one?
Cost per request and quality numbers that settled the choice, not a preference for one approach.
A hiring process for an AI/ML engineer, step by step
Six stages that keep the total short, because good candidates hold competing offers. Stage 5 is where research and production profiles finally separate.
- Pick the profile and the first project Choose research, applied or LLM application, and name the model or feature the hire ships first. Ship support ticket triage with a measured accuracy floor by end of quarter.
- Source from public work Search profiles for production ML evidence, then check Hugging Face, Kaggle or GitHub for the shortlist. Keep data scientists and backend engineers in scope.
- Reach out with the problem and the data ML engineers want to know what data exists and how good it is. Say so in the first message, along with the compute they would get.
- First round on evaluation and serving A 20 minute conversation on shipping, serving, data and cost trade-offs, scored on one rubric for every candidate.
- Paid project or system design A short paid task on a sample of your data, or an ML system design session with your lead. Score the eval plan, not only the model.
- Team round and fast offer Meet the team and make an offer within days of the final round. Delays here cost more than in most roles.
A first round that asks about the eval set
A live, two-way AI video interview that follows up on what the candidate actually says.
In The Cognitive, the rubric is fixed per role and the questions adapt. If a candidate claims they shipped a RAG system, the AI asks what was in the eval set. It picks up to 5 resume claims to probe and marks each verified, refuted or unclear in the report.
How does The Cognitive help you hire an AI/ML engineer?
I run The Cognitive. It handles stages 2 to 4. You paste the brief, search ~900M public profiles, see an "Open to work" badge and tenure per role, and reveal verified emails and phone numbers for the people you want. Then you invite them in bulk to the AI interview for AI/ML engineers, and each books a slot without an account.
Nothing is rejected automatically, and your ML lead reads the reports. If candidates come from your ATS, 60+ integrations write a note back with the score and report link, and stages never move. AI Sourcing and AI Interview each start at $99/month.
More for hiring AI/ML engineers
AI/ML engineer interview questions · AI/ML engineer job description · AI interviewer for AI/ML engineers · Hire a data engineer · AI sourcing tool · Best technical interview software
Frequently asked questions
How much does it cost to hire an AI/ML engineer?
BLS has no ML engineer code. The nearest figures for May 2025 are $120,230 for data scientists, $135,980 for software developers and $140,300 for research scientists. Senior applied roles usually sit above these medians. Tooling is separate: AI Sourcing and AI Interview each start at $99/month.
Where can I find machine learning engineers?
In data science and backend engineering teams as well as ML teams. Public work is the best trail: Hugging Face models and datasets, Kaggle results, GitHub contributions to serving and eval projects, workshop papers and the MLOps Community.
What is the difference between an AI engineer and an ML engineer?
Usage varies. In most job ads, an AI engineer builds features on hosted or open LLMs, while an ML engineer trains, serves and monitors models. Decide which work you need and search for that evidence, since the titles overlap.
How long does it take to hire an AI/ML engineer?
BLS does not publish time to fill for this role. Research roles take longest because the pool is small. For applied roles the risk is losing people to faster processes, so keep the gap between stages short and run the first round as soon as someone replies.
Do I need a PhD for an ML engineer role?
Only for research work. BLS lists a master's degree as typical entry for research scientists. For applied and LLM roles, production evidence such as shipped models, eval sets and monitoring predicts more than a degree.
Can an AI interview assess an ML engineer?
It can assess reasoning and experience: how they evaluated, served and monitored models, and whether resume claims hold up under follow-up questions. It is a spoken conversation, so keep notebooks or code for a later round with your team.
Should I hire a data engineer before an ML engineer?
Often, yes. If your data is scattered or unlabeled, an ML engineer spends the first months building pipelines. Ask candidates how they would check data quality first, and consider whether that work needs its own hire.
Explore: AI/ML engineer interview questions · AI/ML engineer job description · AI interviewer for AI/ML engineers · AI sourcing tool · Pricing