Remote Bengali QA Evaluator — Model Evaluation

AI Trainer Jobs; United States; Remote; USD 15,000 - 21,000; Full time

Job summary

AI Trainer Jobs is hiring Bengali language evaluators to remotely assess model outputs for fluency, accuracy and cultural context. Reviewers provide structured rationale and apply rubrics to guide improvements, working as independent contractors.

Ideal candidates are native or near-native Bengali speakers with strong English, capable of applying rubrics across large evaluation batches, and able to commit at least 10 hours per week.

Qualifications

• Native or near-native Bengali fluency with strong written-English ability.
• Demonstrable experience editing, translating, or evaluating Bengali content.
• Experience applying multi-page rubrics consistently across long evaluation batches.
• Clear written reasoning that names the issue and the standard being applied.
• Reliable async availability for at least 10 hours per week.
• Prior model-evaluation, annotation, or human-rater experience is a plus.

Responsibilities

• Evaluate Bengali model responses for fluency, accuracy, register, and cultural fit for assignments.
• Compare paired Bengali annotation outputs and pick the stronger response with a written rationale.
• Flag hallucinations, code-switching errors, and formality mismatches with task-specific severity tags.
• Capture native-speaker edits that demonstrate the correct phrasing alongside each issue.

Skills

Native Bengali fluency

Strong written English ability

Rubric application

LLM evaluation

Bengali annotation

Quality assessment

Tools

TMS tools

MQM taxonomies

LLM rubrics

Job description

AI Trainer Jobs is hiring Bengali language evaluators to remotely assess model outputs for fluency, accuracy and cultural context. Reviewers provide structured rationale and apply rubrics to guide improvements, working as independent contractors.

Ideal candidates are native or near-native Bengali speakers with strong English, capable of applying rubrics across large evaluation batches, and able to commit at least 10 hours per week.