Esraa Mubarak

Key Skills

Software

Appen

Data Annotation Tech

Google Cloud Vertex AI

Remotasks

Other

Scale AI

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Document

Video

Top Task Types

Action Recognition

Audio Recording

Classification

Emotion Recognition

Evaluation Rating

Freelancer Overview

With hands-on experience in AI training and data annotation, I have contributed to multiple projects focused on Arabic and English language tasks. My work with Sigma AI and Outlier includes prompt evaluation, text classification, translation review, and response ranking for LLMs. In addition to text-based tasks, I provide high-quality audio transcription services in both Arabic and English. I am fluent in Modern Standard Arabic (Fusha) as well as Egyptian, Gulf, and Levantine dialects, allowing me to accurately transcribe and annotate audio data across diverse regions and use cases. I bring a strong understanding of linguistic accuracy, cultural context, and user intent—combined with a technical background in full stack development and IT support. This unique blend enables me to assess model outputs not only linguistically but also functionally. My attention to detail, bilingual proficiency, and adaptability across domains set me apart in delivering accurate, ethical, and scalable training data for AI systems.

Labeling Experience

Arabic Video Content Annotator – Freelance

Internal Proprietary ToolingVideoClassificationEmotion Recognition

The project focused on enhancing the accuracy of Arabic video transcriptions and dubs for AI training datasets. Tasks involved reviewing Arabic video content, identifying and labeling transcription or dubbing errors, classifying them by severity (major/minor), and annotating timestamps. Quality assurance was essential, with strict adherence to grammar, cultural appropriateness, and linguistic accuracy. Worked on multiple video types including educational content, interviews, and AI-generated responses. Annotators followed quality control guidelines and completed feedback loops using tools like Subtitle Edit and Google Docs. The scope of the project included 20+ videos totaling over 5 hours of content reviewed in batches.

The project focused on enhancing the accuracy of Arabic video transcriptions and dubs for AI training datasets. Tasks involved reviewing Arabic video content, identifying and labeling transcription or dubbing errors, classifying them by severity (major/minor), and annotating timestamps. Quality assurance was essential, with strict adherence to grammar, cultural appropriateness, and linguistic accuracy. Worked on multiple video types including educational content, interviews, and AI-generated responses. Annotators followed quality control guidelines and completed feedback loops using tools like Subtitle Edit and Google Docs. The scope of the project included 20+ videos totaling over 5 hours of content reviewed in batches.

2024

LLM Speech Evaluation – Egyptian Arabic (Ongoing)

Internal Proprietary ToolingAudioEmotion RecognitionRLHF

I evaluate audio responses generated by LLMs in Egyptian Arabic. My role involves listening to the model's voice replies to given prompts and scoring them based on accuracy, tone, speed, emotional alignment, and consistency with the given persona. Instructions, platform interface, and feedback loops are provided in English.

I evaluate audio responses generated by LLMs in Egyptian Arabic. My role involves listening to the model's voice replies to given prompts and scoring them based on accuracy, tone, speed, emotional alignment, and consistency with the given persona. Instructions, platform interface, and feedback loops are provided in English.

2024

Voice Recording for AI Training – Egyptian Arabic

Internal Proprietary ToolingAudioAudio Recording

I recorded predefined sentences in Egyptian Arabic with different voice variations based on instructions such as tone, emotion, and speaking speed. The recordings were used to train speech recognition and synthesis models in Arabic.

I recorded predefined sentences in Egyptian Arabic with different voice variations based on instructions such as tone, emotion, and speaking speed. The recordings were used to train speech recognition and synthesis models in Arabic.

2024 - 2024

LLM Response Evaluation – Egyptian Arabic Account (Outlier)

Scale AITextClassificationText Generation

I worked on the Egyptian Arabic account on Outlier to evaluate and compare AI-generated responses in natural language. The tasks included rating and ranking model replies to prompts, providing written explanations for preferences, and assessing fluency, relevance, tone, and clarity of the answers. All responses were in Arabic (Egyptian dialect), and tasks required high linguistic accuracy and a deep understanding of local context and nuance. The goal was to help fine-tune large language models (LLMs) for improved performance in dialectal Arabic.

I worked on the Egyptian Arabic account on Outlier to evaluate and compare AI-generated responses in natural language. The tasks included rating and ranking model replies to prompts, providing written explanations for preferences, and assessing fluency, relevance, tone, and clarity of the answers. All responses were in Arabic (Egyptian dialect), and tasks required high linguistic accuracy and a deep understanding of local context and nuance. The goal was to help fine-tune large language models (LLMs) for improved performance in dialectal Arabic.

2023 - 2024

Education

Work History