About The Role
AI Trainer jobs in Canada focus on improving large language models through RLHF-style evaluations, prompt evaluation, data labeling, and QA evaluation across real AI/ML training pipelines. You will follow detailed annotation guidelines, verify training data quality, and provide structured feedback that improves helpfulness, correctness, and safety.
What You'll Do
- Perform RLHF evaluations (pairwise ranking, rubric-based scoring) and write clear rationales
- Execute prompt evaluation for instruction-following, factuality, and safety
- Label and validate datasets for NLP and content safety labeling
- Run QA evaluation checks (consistency, agreement, systematic error discovery)
- Document edge cases and build error taxonomies to drive model performance improvement
- Collaborate on rubrics, gold sets, calibration, and regression testing for model updates
Skills And Qualifications
- Experience with structured evaluation and guideline-driven judgment
- Strong writing and documentation for rationales and edge-case notes
- Familiarity with RLHF, LLM evaluation, and prompt evaluation workflows
- Comfort with data labeling, QA evaluation, and training data quality processes
- Bonus: multilingual evaluation, NER/classification tasks, or multimodal evaluation
How To Apply
Browse roles on Rex.zone and apply with a resume highlighting RLHF, data labeling, QA evaluation, annotation guidelines compliance, and examples of model evaluation feedback.
Hourly base pay range: $30–$50.
AI Trainer Jobs in Canada in Québec employer: Rex.zone
Join a forward-thinking team in Canada dedicated to advancing AI technology through innovative training methodologies. As an AI Trainer, you'll benefit from a collaborative work culture that prioritizes employee growth and development, offering opportunities to enhance your skills in a rapidly evolving field. With competitive pay and a focus on meaningful contributions to AI safety and effectiveness, this role is perfect for those looking to make a significant impact in the tech industry.
