Yaroslav Shyryayev

Key Skills

Software

Appen

Labelbox

Mercor

OneForma

Scale AI

SuperAnnotate

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Text

Top Label Types

RLHF

Red Teaming

Evaluation Rating

Prompt Response Writing SFT

Entity Ner Classification

Freelancer Overview

I am a multilingual AI Data Specialist with extensive experience in data labeling, annotation, and AI training data for leading platforms such as Outlier, Alignerr, Appen, and OneForma. My work has focused on evaluating and refining Large Language Models (LLMs) through Reinforcement Learning from Human Feedback (RLHF), prompt engineering, red teaming, and fact-checking, ensuring high-quality, safe, and aligned AI outputs. I am skilled in generating and annotating complex prompts across Italian, Russian, and English, and have a strong background in both NLP and medical data workflows, including digital transformation using CAD/CAM and 3D printing technologies. My technical toolkit includes SQL, advanced Excel, ChatGPT, Claude, Copilot, and various annotation interfaces. I am highly organized, detail-oriented, and certified in advanced AI prompting, data analysis, and agile project management, enabling me to deliver accurate, reliable, and efficient training data for cutting-edge AI applications.

Labeling Experience

AI Red Teaming & Safety Specialist

SuperannotateTextRLHFRed Teaming

Adversarial testing and safety alignment for Large Language Models, specifically tailored for the Italian language market. Designing complex "jailbreak" prompts to bypass safety filters, identifying biases, and testing model boundaries regarding prohibited content, ethical guidelines, and security.Daily high-intensity testing sessions targeting various safety taxonomies. Adherence to strict "Harmlessness" policies and daily calibration with the safety team to ensure zero-tolerance for unsafe model outputs.

Adversarial testing and safety alignment for Large Language Models, specifically tailored for the Italian language market. Designing complex "jailbreak" prompts to bypass safety filters, identifying biases, and testing model boundaries regarding prohibited content, ethical guidelines, and security.Daily high-intensity testing sessions targeting various safety taxonomies. Adherence to strict "Harmlessness" policies and daily calibration with the safety team to ensure zero-tolerance for unsafe model outputs.

2025 - 2024

LLM Evaluation

LabelboxTextRLHF

Improving model reasoning and helpfulness through Reinforcement Learning from Human Feedback (RLHF).Crafting complex prompts and performing side-by-side (Model A vs Model B) evaluations. Ranking responses based on accuracy, tone, and formatting.Maintained a high Quality Score (QA) through rigorous fact-checking and adherence to project-specific style guides.

Improving model reasoning and helpfulness through Reinforcement Learning from Human Feedback (RLHF).Crafting complex prompts and performing side-by-side (Model A vs Model B) evaluations. Ranking responses based on accuracy, tone, and formatting.Maintained a high Quality Score (QA) through rigorous fact-checking and adherence to project-specific style guides.

2024 - 2025

Complex Audio Transcription & Sound Identification

AppenAudioEntity Ner Classification

Developing datasets for advanced speech recognition and sound detection.Transcribing verbatim audio files and inserting specific tags (Entity classification) for non-verbal sounds (sighs, pauses, background noise).Processed hundreds of hours of multi-speaker audio. Adherence to strict transcription conventions and time-stamping accuracy within millisecond thresholds.

Developing datasets for advanced speech recognition and sound detection.Transcribing verbatim audio files and inserting specific tags (Entity classification) for non-verbal sounds (sighs, pauses, background noise).Processed hundreds of hours of multi-speaker audio. Adherence to strict transcription conventions and time-stamping accuracy within millisecond thresholds.

2024 - 2024

LLM Training & Comparative Analysis

AppenTextPrompt Response Writing SFT

Training Generative AI models to provide better creative and logical responses.Writing original prompts to trigger specific model behaviors and evaluating the quality of two competing model outputs.

Training Generative AI models to provide better creative and logical responses.Writing original prompts to trigger specific model behaviors and evaluating the quality of two competing model outputs.

2024 - 2024

Advanced LLM Response Evaluation (RLHF)

OneformaTextEvaluation Rating

Improving model reasoning and helpfulness through Reinforcement Learning from Human Feedback (RLHF).

Improving model reasoning and helpfulness through Reinforcement Learning from Human Feedback (RLHF).

2024 - 2024

Education

Work History