Saeto Kanda

Key Skills

Software

Appen

OneForma

Scale AI

Don't disclose

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Label Types

Classification

Evaluation Rating

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I am passionate about leveraging my strong analytical and problem-solving skills in the field of AI and data labeling. With hands-on experience in grading academic tests and providing personalized tutoring in subjects such as mathematics, English, and Japanese, I have developed a keen eye for detail, accuracy, and fairness—qualities essential for high-quality data annotation. My background in Japanese writing, editing, and proofreading enables me to deliver precise and reliable language data, while my proficiency in English allows me to work effectively in multilingual environments. I am eager to contribute to projects in natural language processing and educational data, utilizing my skills to support the development of innovative AI solutions.

Labeling Experience

Native Japanese Audio Transcriber – Mixed Script

LabelboxAudioEvaluation RatingTranscription

This project involved transcribing short Japanese audio clips for AI training, following strict style and annotation guidelines. I produced accurate mixed-script transcriptions (Kanji, Hiragana, Katakana), handled multi-speaker audio, captured required speaker metadata, selected appropriate save states with brief reasoning, and submitted high-quality work within a QC-based review workflow.

This project involved transcribing short Japanese audio clips for AI training, following strict style and annotation guidelines. I produced accurate mixed-script transcriptions (Kanji, Hiragana, Katakana), handled multi-speaker audio, captured required speaker metadata, selected appropriate save states with brief reasoning, and submitted high-quality work within a QC-based review workflow.

2025 - 2025

English-to-Japanese Translation Review and Correction for AI Systems

Don T DiscloseTextText GenerationEvaluation Rating

In this project, I reviewed and corrected Japanese translations generated by AI from English source texts across a wide range of topics. Tasks included evaluating the accuracy and fluency of the translations, identifying and fixing mistranslations, improving unnatural phrasing, and ensuring consistency in tone and terminology. The work involved side-by-side comparison of English originals and AI-generated Japanese outputs, with revisions made according to detailed linguistic and stylistic guidelines. Each submission was subject to quality review by multiple annotators, and performance was scored regularly. I committed approximately 20 hours per week to this project, maintaining high quality standards throughout.

In this project, I reviewed and corrected Japanese translations generated by AI from English source texts across a wide range of topics. Tasks included evaluating the accuracy and fluency of the translations, identifying and fixing mistranslations, improving unnatural phrasing, and ensuring consistency in tone and terminology. The work involved side-by-side comparison of English originals and AI-generated Japanese outputs, with revisions made according to detailed linguistic and stylistic guidelines. Each submission was subject to quality review by multiple annotators, and performance was scored regularly. I committed approximately 20 hours per week to this project, maintaining high quality standards throughout.

2025 - 2025

Japanese Audio Data Collection and Transcription for ASR Training

Scale AIAudioAudio Recording

In this project, I participated in the collection and transcription of short speech recordings (approximately 30 seconds each) in Japanese, aimed at improving automatic speech recognition (ASR) systems. The recordings were based on both free-form natural conversations and predefined scenarios, captured in a wide range of real-world environments using smartphones. Over 1,000 audio samples were submitted, and each was subject to multi-stage quality checks and peer feedback. Strict evaluation standards were enforced—participants whose submissions consistently fell below the quality threshold were removed from the project. I dedicated roughly 10 hours per week to this task and maintained high accuracy throughout. This project required not only clear and consistent speech delivery, but also attention to environmental sound conditions and detailed adherence to guidelines.

In this project, I participated in the collection and transcription of short speech recordings (approximately 30 seconds each) in Japanese, aimed at improving automatic speech recognition (ASR) systems. The recordings were based on both free-form natural conversations and predefined scenarios, captured in a wide range of real-world environments using smartphones. Over 1,000 audio samples were submitted, and each was subject to multi-stage quality checks and peer feedback. Strict evaluation standards were enforced—participants whose submissions consistently fell below the quality threshold were removed from the project. I dedicated roughly 10 hours per week to this task and maintained high accuracy throughout. This project required not only clear and consistent speech delivery, but also attention to environmental sound conditions and detailed adherence to guidelines.

2025 - 2025

News Article Data Labeler

AppenDocumentClassificationEvaluation Rating

This project involved annotating and evaluating Japanese-language news articles for AI training. The work focused on labeling article structure and content types—such as headlines, factual statements, quotations, and opinionated content—while assessing clarity, consistency, and linguistic naturalness based on standardized guidelines.

This project involved annotating and evaluating Japanese-language news articles for AI training. The work focused on labeling article structure and content types—such as headlines, factual statements, quotations, and opinionated content—while assessing clarity, consistency, and linguistic naturalness based on standardized guidelines.

2024 - 2024

Human-Like Japanese Output Review and Correction

Don T DiscloseTextRLHF

In this RLHF project, I evaluated and refined Japanese outputs generated by a language model, with the goal of making them more natural, human-like, and contextually appropriate. Beyond checking for grammatical accuracy, I focused on tone, clarity, coherence, and whether the response aligned with the intended instruction or user prompt. Following detailed guidelines, I provided feedback and rewrote outputs when necessary to improve fluency and user-friendliness. All submissions underwent strict quality control, including multi-level reviews and regular scoring. I consistently maintained high performance standards while contributing approximately 20 hours per week.

In this RLHF project, I evaluated and refined Japanese outputs generated by a language model, with the goal of making them more natural, human-like, and contextually appropriate. Beyond checking for grammatical accuracy, I focused on tone, clarity, coherence, and whether the response aligned with the intended instruction or user prompt. Following detailed guidelines, I provided feedback and rewrote outputs when necessary to improve fluency and user-friendliness. All submissions underwent strict quality control, including multi-level reviews and regular scoring. I consistently maintained high performance standards while contributing approximately 20 hours per week.

2024 - 2024

Education

Work History