Özgür Öztürk

Key Skills

Software

Label Studio

LightTag

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Text

Top Task Types

Prompt Response Writing SFT

Question Answering

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I bring a unique blend of linguistic expertise and technical competency to AI training data tasks, particularly in the evaluation and refinement of large language model (LLM) outputs. As a certified translator fluent in Turkish, English, Spanish, and Russian, I have a deep understanding of semantic nuance, syntactic structure, and contextual variation across languages. This enables me to critically assess multilingual LLM responses for accuracy, coherence, tone, and cultural relevance—key factors in reducing hallucinations and improving user experience in generative AI systems. My technical toolkit includes Python (Pandas, NumPy), SQL, and data visualization tools like Tableau and Power BI, which I've used in analytical projects involving text data, campaign performance metrics, and prompt evaluation workflows. I’ve also applied structured QA methodologies and A/B testing to assess chatbot and text generation output in SaaS and customer support environments. With strong attention to linguistic detail and a working knowledge of machine learning fundamentals, I’m well-positioned to contribute to LLM evaluation, prompt tuning, and data quality audits at scale.

Labeling Experience

Video Action Recognition Annotation for Human Behavior Detection

LabelboxVideoBounding BoxPolygon

Worked on a large-scale video annotation project involving the labeling of human actions and object interactions across surveillance-style video datasets. Tasks included drawing bounding boxes and polygons around people and objects, tagging actions (e.g., walking, sitting, waving), and using object tracking tools to maintain label consistency across frames. I used CVAT’s interpolation feature and Labelbox’s collaborative QA system to ensure efficiency. Maintained an annotation accuracy above 98% according to internal QA metrics. The dataset contributed to training models for a real-time behavior recognition system used in a smart city pilot.

Worked on a large-scale video annotation project involving the labeling of human actions and object interactions across surveillance-style video datasets. Tasks included drawing bounding boxes and polygons around people and objects, tagging actions (e.g., walking, sitting, waving), and using object tracking tools to maintain label consistency across frames. I used CVAT’s interpolation feature and Labelbox’s collaborative QA system to ensure efficiency. Maintained an annotation accuracy above 98% according to internal QA metrics. The dataset contributed to training models for a real-time behavior recognition system used in a smart city pilot.

2024 - 2024

Audio Data Labeling Specialist

LighttagAudioEmotion Recognition

Labeled short audio clips in Turkish and English for emotional tone, speaker turns, and intent. Helped improve the accuracy of speech-based AI by providing human judgment in edge cases and ambiguous inputs.

Labeled short audio clips in Turkish and English for emotional tone, speaker turns, and intent. Helped improve the accuracy of speech-based AI by providing human judgment in edge cases and ambiguous inputs.

2024 - 2024

Multilingual LLM Response Evaluation and Prompt QA

Label StudioTextText Generation

Evaluated multilingual LLM-generated responses (Turkish, English, Spanish, Russian) for semantic accuracy, cultural relevance, fluency, and contextual appropriateness. Used structured rating systems to assess hallucinations, tone consistency, and factual grounding across customer service and general knowledge prompts. Conducted manual annotations for NER tasks and participated in A/B testing of prompt variations to enhance generative reliability. Familiar with post-editing machine translation outputs and refining data quality for training purposes.

Evaluated multilingual LLM-generated responses (Turkish, English, Spanish, Russian) for semantic accuracy, cultural relevance, fluency, and contextual appropriateness. Used structured rating systems to assess hallucinations, tone consistency, and factual grounding across customer service and general knowledge prompts. Conducted manual annotations for NER tasks and participated in A/B testing of prompt variations to enhance generative reliability. Familiar with post-editing machine translation outputs and refining data quality for training purposes.

2024 - 2024

Education

Work History