About The Role
As a Germany-based English & German AI Generalist Trainer at Rex.zone, you will evaluate and improve AI systems by reviewing model-generated outputs in real-world scenarios. Your work supports RLHF, large language model evaluation, and consistent training data quality through careful labeling, prompt evaluation, and annotation guidelines compliance.
About The Role
As a Germany-based English & German AI Generalist Trainer at Rex.zone, you will evaluate and improve AI systems by reviewing model-generated outputs in real-world scenarios. Your work supports RLHF, large language model evaluation, and consistent training data quality through careful labeling, prompt evaluation, and annotation guidelines compliance.
Key Responsibilities
- Evaluate and rank model-generated responses against defined rubrics and project goals
- Perform QA evaluation by validating labels, rationales, and final outputs for accuracy and consistency
- Write concise reasoning and justifications in German and English to support RLHF datasets
- Apply annotation guidelines compliance and flag ambiguity or guideline gaps for resolution
- Execute prompt evaluation and large language model evaluation across multiple task types
- Conduct validation checks for content safety labeling and policy adherence
- Identify recurring failure modes and suggest improvements that increase training data quality and model performance
Basic Qualifications
- Must be based in Germany (remote role) and authorized to work as applicable
- Fluent in German and English (reading and writing) with strong grammar and style
- Strong analytical skills, attention to detail, and ability to follow multi-step rubrics
- Ability to document clear, consistent rationales
- Reliable internet connection and ability to meet quality and productivity targets
Preferred Qualifications
- Experience with data labeling, QA evaluation, or annotation projects
- Familiarity with LLMs, RLHF concepts, and evaluation workflows
- Background in linguistics, translation, writing, journalism, customer support, or technical documentation
#J-18808-Ljbffr