Celine Castex

Key Skills

Software

Data Annotation Tech

Labelbox

Other

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Document

Text

Top Task Types

Audio Recording

Classification

Evaluation Rating

Prompt Response Writing SFT

Translation Localization

Freelancer Overview

As a dedicated AI Trainer and Bilingual Content Editor, I specialize in enhancing and refining Artificial Intelligence models, with a particular focus on French-language content. My career, spanning over a decade in high-precision linguistic roles (translation, interpreting, editing), has cultivated a unique skillset in critical analysis, linguistic nuance, and meticulous data evaluation—all directly applicable to advanced AI training. My current role as a Bilingual Content Editor - AI Trainer directly involves rigorous evaluation and refinement of AI-generated text, ensuring its factual accuracy, grammatical correctness, and cultural relevance. I am adept at producing high-quality original content for model benchmarks and providing insightful feedback to improve AI performance and mitigate biases. With a strong foundation in both French (native from France) and English, coupled with diverse cross-sector experience—from complex medical documentation to sensitive international relations—I bring an unparalleled understanding of context and subject matter. This versatile background enables me to identify subtle linguistic errors and cultural inconsistencies that are critical for developing robust and ethical AI systems. I am passionate about contributing to the cutting edge of AI, leveraging my expertise to drive improvements in NLP and LLMs. I am eager to apply my proven ability to analyze, refine, and optimize linguistic data to shape the next generation of intelligent AI.

Labeling Experience

Bilingual AI Trainer (French)

Data Annotation TechTextEvaluation Rating

his project involves evaluating AI model responses based on specific rating axes: Verbosity, Instruction Following, Truthfulness, and Overall Quality. The core task is to assess if the AI provides the right amount of information (Verbosity), accurately follows all prompt instructions (Instruction Following), delivers factually correct information (Truthfulness), and presents a high-quality overall output. A key aspect is understanding subtle distinctions between these axes and providing detailed, concise comments for any identified issues, often requiring "best judgment" within clear guidelines. The goal is to provide targeted feedback to improve AI model performance.

his project involves evaluating AI model responses based on specific rating axes: Verbosity, Instruction Following, Truthfulness, and Overall Quality. The core task is to assess if the AI provides the right amount of information (Verbosity), accurately follows all prompt instructions (Instruction Following), delivers factually correct information (Truthfulness), and presents a high-quality overall output. A key aspect is understanding subtle distinctions between these axes and providing detailed, concise comments for any identified issues, often requiring "best judgment" within clear guidelines. The goal is to provide targeted feedback to improve AI model performance.

2024

Bilingual Editor (French) AI Trainer for Outlier

Internal Proprietary ToolingTextEvaluation Rating

Engaged in critical AI safety missions on Outlier, meticulously crafting prompts on sensitive topics (e.g., self-harm, politics). My role involved rigorously evaluating AI responses for appropriateness and bias, directly contributing to the development of safer and more ethical AI systems. This mission, focused on "Safety," was a critical component of Responsible AI development, specifically designed to evaluate and mitigate harmful or inappropriate outputs from AI models. The core objective was to stress-test AI systems by prompting them with challenging and sensitive topics to ensure their responses were appropriate, harmless, and unbiased.

Engaged in critical AI safety missions on Outlier, meticulously crafting prompts on sensitive topics (e.g., self-harm, politics). My role involved rigorously evaluating AI responses for appropriateness and bias, directly contributing to the development of safer and more ethical AI systems. This mission, focused on "Safety," was a critical component of Responsible AI development, specifically designed to evaluate and mitigate harmful or inappropriate outputs from AI models. The core objective was to stress-test AI systems by prompting them with challenging and sensitive topics to ensure their responses were appropriate, harmless, and unbiased.

2024

Bilingual Editor (French)- AI Trainer

Data Annotation TechAudioAudio Recording

This mission focused on collecting diverse audio data to train and improve AI models, particularly in the realm of speech recognition and natural language understanding (NLU) in challenging acoustic environments. The objective was to create a dataset that teaches AI to accurately interpret user requests even when background noise is present. Key Components and Objectives: Audio Recording (Under 1 Minute): Purpose: To capture short, distinct audio samples that represent typical user interactions with AI assistants or voice-controlled systems. The sub-1-minute duration is optimal for capturing concise commands, questions, or short conversational turns, and for efficient processing in AI training. Content: The recorded audio should comprise "requests AI can respond to." This means the audio should be natural language prompts, commands, or questions that a user would typically ask a virtual assistant, smart speaker, or other voice-enabled AI system. Examples could include: "Play some ja

This mission focused on collecting diverse audio data to train and improve AI models, particularly in the realm of speech recognition and natural language understanding (NLU) in challenging acoustic environments. The objective was to create a dataset that teaches AI to accurately interpret user requests even when background noise is present. Key Components and Objectives: Audio Recording (Under 1 Minute): Purpose: To capture short, distinct audio samples that represent typical user interactions with AI assistants or voice-controlled systems. The sub-1-minute duration is optimal for capturing concise commands, questions, or short conversational turns, and for efficient processing in AI training. Content: The recorded audio should comprise "requests AI can respond to." This means the audio should be natural language prompts, commands, or questions that a user would typically ask a virtual assistant, smart speaker, or other voice-enabled AI system. Examples could include: "Play some ja

2024

Bilingual AI Trainer (French) for Outlier

Internal Proprietary ToolingTextPrompt Response Writing SFT

1. Rubric Application: Rapidly applying complex evaluation rubrics to assess AI-generated content (e.g., factuality, coherence, grammar, style, safety). 2. Comparative Analysis: Quickly comparing multiple AI responses to a single prompt and ranking them based on specified criteria. 3. Error Identification: Efficiently spotting various types of errors (grammatical, factual, logical, stylistic) within AI text. 4. Content Generation (Concise): Producing short, high-quality original content in response to prompts under time constraints. 5. Bias Detection: Identifying instances of bias or harmful content in AI outputs. 6. Instruction Following: Precisely adhering to detailed and sometimes complex project guidelines. 7. Domain-Specific Evaluation: If the course was specialized, it might involve tasks related to specific domains (e.g., medical, legal, technical, creative writing) to test subject matter expertise under pressure.

1. Rubric Application: Rapidly applying complex evaluation rubrics to assess AI-generated content (e.g., factuality, coherence, grammar, style, safety). 2. Comparative Analysis: Quickly comparing multiple AI responses to a single prompt and ranking them based on specified criteria. 3. Error Identification: Efficiently spotting various types of errors (grammatical, factual, logical, stylistic) within AI text. 4. Content Generation (Concise): Producing short, high-quality original content in response to prompts under time constraints. 5. Bias Detection: Identifying instances of bias or harmful content in AI outputs. 6. Instruction Following: Precisely adhering to detailed and sometimes complex project guidelines. 7. Domain-Specific Evaluation: If the course was specialized, it might involve tasks related to specific domains (e.g., medical, legal, technical, creative writing) to test subject matter expertise under pressure.

2024

Education

Work History