Fumiyo Kumagai

Key Skills

Software

Other

Labelbox

Mercor

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Text

Top Label Types

Segmentation

Evaluation Rating

Data Collection

Audio Recording

Transcription

Classification

Red Teaming

Freelancer Overview

I am a native Japanese AI language and cultural evaluator with hands-on experience in AI data labeling, linguistic annotation, and LLM safety review for global training projects. My background includes evaluating model outputs for policy compliance, content safety, and cultural appropriateness, with a strong focus on nuanced prompt assessment and toxicity annotation. I have worked on high-volume annotation tasks for platforms such as TaskUs and OpentrainAI, consistently applying complex English guidelines to Japanese content and providing clear, structured justifications for safety and cultural judgments. My expertise extends to audio transcription, scriptwriting, and narrative design, which supports my strong contextual understanding and natural language judgment. Bilingual in Japanese and English, I am adept at ensuring data quality, accuracy, and cultural relevance across diverse AI and NLP projects.

Labeling Experience

LingoTribe – Japanese Audio Transcription and Voice Recording for Conversational AI

OtherAudioSegmentationEvaluation Rating

Worked on large-scale Japanese conversational AI training projects through transcription and audio recording tasks. Responsibilities included accurately transcribing natural spoken dialogues, segmenting multiple speakers, and evaluating recording quality using external microphones. Ensured linguistic precision, cultural nuance, and adherence to style and accuracy guidelines (over 96% average accuracy). Contributed to improving AI understanding of natural Japanese speech through detailed annotation and quality feedback.

Worked on large-scale Japanese conversational AI training projects through transcription and audio recording tasks. Responsibilities included accurately transcribing natural spoken dialogues, segmenting multiple speakers, and evaluating recording quality using external microphones. Ensured linguistic precision, cultural nuance, and adherence to style and accuracy guidelines (over 96% average accuracy). Contributed to improving AI understanding of natural Japanese speech through detailed annotation and quality feedback.

2025 - 2025

TaskUs (Cinder) – Japanese Content Safety & Abusive Language Annotation

OtherTextClassificationEvaluation Rating

Labeled and evaluated Japanese user-generated content for abusive, discriminatory, and policy-sensitive language. Assessed contextual meaning, implied intent, and severity levels based on detailed safety guidelines. Maintained consistency across high-volume tasks with frequent guideline updates.

Labeled and evaluated Japanese user-generated content for abusive, discriminatory, and policy-sensitive language. Assessed contextual meaning, implied intent, and severity levels based on detailed safety guidelines. Maintained consistency across high-volume tasks with frequent guideline updates.

2025

Mercor – AI Safety Red Teaming (Work Trial – Passed)

MercorTextClassificationEvaluation Rating

Completed and passed an AI Safety Red Teaming work trial. Reviewed adversarial and policy-sensitive prompts and evaluated AI responses for safety risks. Provided structured justification for decisions in English based on provided guidelines.

Completed and passed an AI Safety Red Teaming work trial. Reviewed adversarial and policy-sensitive prompts and evaluated AI responses for safety risks. Provided structured justification for decisions in English based on provided guidelines.

2025 - 2025

OpenTrain AI – Japanese Audio Transcription & Text Annotation

LabelboxAudioSegmentationEvaluation Rating

Performed Japanese audio transcription and text annotation tasks following detailed English guidelines. Focused on accuracy, consistency, and linguistic quality checks for AI training datasets. Completed high-volume labeling tasks and met project quality requirements.

Performed Japanese audio transcription and text annotation tasks following detailed English guidelines. Focused on accuracy, consistency, and linguistic quality checks for AI training datasets. Completed high-volume labeling tasks and met project quality requirements.

2025 - 2025

Education

Work History