AI Training Jobs for Data Scientists and ML Engineers

Yes, data scientists and machine learning engineers can do remote AI training work, and they often have an advantage: they understand what the labels, rubrics and comparisons are actually for. Roles in this area tend to focus on evaluation, data quality and technical review rather than on building models. This guide covers the kinds of tasks listed, what pay figures show, and how to present yourself honestly.

What "training" means for someone who builds models

If you build models for a living, it helps to separate two things. Platforms such as Mercor and Micro1 do not usually hire you to train a network on your own GPU. They match professionals with AI labs and companies that need human input for training, evaluation and data work. That input could be:

• Evaluation design. Writing test cases, rubrics and reference answers that measure whether a model reasons correctly in statistics, ML, data analysis or code.
• Grading and ranking. Comparing model outputs on technical questions and explaining which is better. This is the human feedback step in RLHF; see what RLHF is and what AI trainers actually do if the term is new.
• Code and notebook review. Checking generated Python, SQL or analysis code for correctness, efficiency and bad statistical practice.
• Fact-checking technical explanations. Spotting confident but wrong claims about algorithms, metrics, leakage or experimental design.
• Data quality work. Reviewing, cleaning or labelling datasets with domain judgement that non-specialists lack.
• Red-teaming. Probing models for failure modes, in some projects. Our guide on getting into AI red-teaming goes deeper.

The skill you bring is not just knowing the answer. It is noticing why a plausible-sounding answer is wrong, which is a daily habit for anyone who has debugged a model.

What listings show

At the time of writing (Oct 2026), SOJI had 32 active jobs in the AI/ML and data science category, with a typical median listed minimum of about $60/h and a highest listed maximum of up to $350/h. For comparison, software engineering had 53 jobs with a median minimum around $50/h. These figures are what listings display, not what people earn, and they vary a lot by project, seniority and whether a role pays per task.

Category	Active jobs	Typical median of listed minimum
AI/ML and data science	32	about $60/h
Software engineering	53	about $50/h
Engineering and science (maths, physics)	46	about $80/h

Do not read the table as a ranking of what your skills are worth. Rates are tied to project demand at a given moment, and seniority matters. If a listing says per task, work out your effective hourly rate with our guide to comparing hourly and per-task pay.

How to position yourself

1. Lead with the specific subfield. "NLP and evaluation metrics", "time series forecasting", "causal inference", "computer vision data pipelines" are more useful to a matcher than "data scientist". Name languages and libraries you really use.
2. Show evidence of rigour. Published work, open-source contributions, well-documented projects and rigorous experimentation are relevant. Be accurate about your role in team projects.
3. Do not oversell AI knowledge. Honest answers about what you have and have not done are better than inflated ones. If an interview asks about your field, answer in specifics and say when you are unsure.
4. Mind the two paths. If you are closer to engineering, the software engineering jobs and our post on AI training jobs for software engineers may fit better. If you are closer to research or maths, see AI training jobs for engineers and scientists.
5. Check the resume basics. Our guide on writing a resume for AI training jobs without AI experience is aimed at other fields, but the point about translating experience into concrete outcomes applies equally to you.

The sequence on Mercor is typically: profile and resume, a domain-expert interview about your field, a work-authorization and location check, then matching to a project. See how getting hired on Mercor works step by step. On Micro1, follow the steps on each listing. In every case, give honest answers; do not use outside help to answer assessments on your behalf.

Eligibility and screening

Every SOJI job page shows "Eligible locations". Roughly 4 in 5 listings are open worldwide, while roughly 1 in 5 are restricted to specific countries or regions, most often the United States. A few roles are on-site or hybrid. Screening questions commonly seen in listings include expected hourly rate in USD, how soon you can start, hours per week available (some roles ask for 10 to 15 or more), and years of experience with a skill. Some roles are paid per task.

Applying to several relevant listings is fine, since screening steps are often shared. Matching is not guaranteed, and some people are matched within days while others wait weeks.

Things worth thinking about before you apply

• Conflict of interest. If your employer builds or competes with AI models, check your contract and any policy on outside work. A job board cannot answer this for you.
• Confidential material. Never use employer code, data or internal documents in an evaluation task.
• Is it a good use of your time? If your day-job salary is high, an hourly gig has to compete with your time. For some people the appeal is variety or flexibility rather than the rate; our article on whether AI training jobs are worth it weighs this.
• Contractor status. Engagements are typically independent-contractor, so taxes and benefits are your responsibility. This is general information, not tax advice; check your country's rules or ask an accountant.

Quick checklist

• You can name your subfield, tools and years in one sentence.
• You have checked "Eligible locations" and the weekly hours expected.
• You know whether pay is hourly or per task.
• You have reviewed your employment contract for outside-work and IP clauses.
• Your resume describes real outcomes, not buzzwords.
• The listing is still open on the platform.

Limitations

These roles are not model-building jobs, and some readers expect research or engineering work and find evaluation tasks repetitive. Listed pay is not a promise of income, hours are often variable and being hired is never guaranteed. The market also shifts as labs change what they need. Start with the AI/ML and data science listings and see all current options at /jobs, and keep expectations measured.

SOJI is an independent job board and is not affiliated with Mercor or Micro1. If you apply through our referral links and are hired, SOJI may earn a commission, at no cost to you, and this does not affect which jobs we publish. Learn more on our about page.