Takaaki Meguro

Key Skills

Software

Data Annotation Tech

Labelbox

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Document

Image

Top Task Types

Action Recognition

Evaluation/Rating

Prompt + Response Writing (SFT)

Red Teaming

RLHF

Freelancer Overview

I am a bilingual AI Data Quality Controller and LLM Evaluator with a focus on text-based model evaluation, instruction tuning, and prompt optimization. My recent work with Outlier AI involved assessing large-scale model outputs for factual accuracy, contextual alignment, and stylistic coherence across English and Japanese datasets. I have a strong understanding of evaluation rubrics, data annotation workflows, and multilingual linguistic nuances, enabling me to deliver high-quality insights that improve model performance and user trust.

Labeling Experience

Cypher Human Evals & RLHF

Scale AITextRLHFFine Tuning

Compared and rated model-generated responses based on multiple evaluation criteria, including localization, instruction following, truthfulness, and style consistency. Contributed to RLHF by crafting and rewriting prompts to improve model alignment with human intent. Delivered detailed qualitative feedback to enhance overall model coherence and user experience.

Compared and rated model-generated responses based on multiple evaluation criteria, including localization, instruction following, truthfulness, and style consistency. Contributed to RLHF by crafting and rewriting prompts to improve model alignment with human intent. Delivered detailed qualitative feedback to enhance overall model coherence and user experience.

2024

Tree_street_rating

Scale AIAudioEmotion RecognitionEvaluation Rating

Conducted evaluation of recorded audio clips by annotating detailed attributes such as gender, age, tone, and speaking style. Tasks included transcribing speech content, identifying situational context, and rating sound quality levels. Focused on ensuring accurate metadata labeling and consistent evaluation standards for multilingual speech datasets.

Conducted evaluation of recorded audio clips by annotating detailed attributes such as gender, age, tone, and speaking style. Tasks included transcribing speech content, identifying situational context, and rating sound quality levels. Focused on ensuring accurate metadata labeling and consistent evaluation standards for multilingual speech datasets.

2025 - 2025

i18n_gauntlet_adversarial_safety

Scale AIDocumentEvaluation RatingRed Teaming

Performed adversarial red teaming to assess AI model responses for safety and ethical compliance. Reviewed text, image, and code-based outputs to detect harmful or unsafe content and ensure proper refusal behavior. Evaluated model robustness across multilingual and multimodal prompts to strengthen safety alignment and reduce risk exposure.

Performed adversarial red teaming to assess AI model responses for safety and ethical compliance. Reviewed text, image, and code-based outputs to detect harmful or unsafe content and ensure proper refusal behavior. Evaluated model robustness across multilingual and multimodal prompts to strengthen safety alignment and reduce risk exposure.

2024 - 2025

Education

Work History