Remote STEM Jobs in Canada

Remote STEM Jobs in Canada (Full Time)

Rex.zone connects mid-senior engineers and STEM professionals to real-world AI/ML training workflows, including LLM evaluation, RLHF-style preference ranking, data labeling, QA evaluation, and prompt evaluation. You will help improve model performance by producing and reviewing high-quality training data and enforcing annotation guidelines compliance.

What You Will Do

Contribute to training data quality through labeling, review, and adjudication
Perform RLHF-style preference ranking and helpfulness/harmlessness evaluations
Execute prompt evaluation and response grading for large language model evaluation
Apply annotation guidelines, document edge cases, and support rubric adherence
Run QA evaluation workflows, track defects, and recommend process improvements
Support NLP tasks (e.g., named entity recognition, taxonomy tagging)
Support computer vision annotation (e.g., bounding boxes, polygons, classification)
Support content safety labeling (policy categories, risk scoring, refusals)
Collaborate with teams across AI labs, tech startups, annotation vendors, and BPO operations

Required Qualifications

Mid-senior experience in STEM or engineering
Strong analytical writing and attention to detail for evaluation rubrics
Familiarity with AI/ML concepts, LLM behavior, and model failure modes
Experience with data labeling, QA evaluation, or guideline-driven review
Ability to work full-time remotely with reliable internet and secure work practices

Preferred Qualifications

Exposure to RLHF, prompt evaluation, and rubric-based grading
Experience with NLP and/or computer vision annotation
Experience with content safety labeling and policy enforcement
Comfort using annotation platforms, spreadsheets, and issue trackers
Ability to mentor peers on annotation guidelines compliance and training data quality

How To Apply

Apply via Rex.zone and highlight your STEM/engineering background, guideline-driven work, and examples that improved training data quality or model performance.
Show more Show less