Samuel Olubunmi

Key Skills

Software

Labelbox

Other

Top Subject Matter

Robotics Domain Expertise

Multimodal Evaluation

Hallucination detection + response ranking

Top Data Types

Video

Text

Image

Top Task Types

Fine-tuning

Classification

Bounding Box

Entity (NER) Classification

Text Generation

Text Summarization

RLHF

Transcription

Evaluation/Rating

Segmentation

Polygon

Data Collection

Question Answering

Freelancer Overview

LLM Trainer and AI Data Quality Analyst with experience in multimodal annotation, NLP, and computer vision datasets. Skilled in RLHF evaluation, prompt-response analysis, and hallucination detection, with 93–98% accuracy across 10,000+ annotations. Proficient in CVAT, Labelbox, and AI data QA workflows.

Labeling Experience

Quality Analyst

Text

As a Quality Analyst, I reviewed and refined labeled data across text, image, audio, and video formats for AI model training. I provided feedback to optimize annotation tools and helped improve annotation processes. My role included segmenting and categorizing data to ensure high dataset quality. • Conducted QA checks and batch evaluations for annotation teams. • Enhanced workflow productivity via detailed tool/process feedback. • Collaborated with data scientists to resolve label ambiguities. • Regularly improved dataset integrity and annotation accuracy.

As a Quality Analyst, I reviewed and refined labeled data across text, image, audio, and video formats for AI model training. I provided feedback to optimize annotation tools and helped improve annotation processes. My role included segmenting and categorizing data to ensure high dataset quality. • Conducted QA checks and batch evaluations for annotation teams. • Enhanced workflow productivity via detailed tool/process feedback. • Collaborated with data scientists to resolve label ambiguities. • Regularly improved dataset integrity and annotation accuracy.

2024 - Present

LLM Trainer: Vision-Language-Action and Video Game Captioning

LabelboxVideoFine Tuning

As an LLM Trainer, I annotated multimodal datasets to support robotics and video game AI research. I utilized Chain-of-Thought reasoning to provide structured annotations across vision, language, and action sequences. Quality and linguistic precision were emphasized to exceed benchmark QA requirements. • Annotated 2,000+ samples for LLM fine-tuning across robotic and video game domains. • Applied reasoning to capture complex temporal and action relationships. • Supported embodied AI models through high-accuracy annotation. • Consistently achieved a 98% QA accuracy rate.

As an LLM Trainer, I annotated multimodal datasets to support robotics and video game AI research. I utilized Chain-of-Thought reasoning to provide structured annotations across vision, language, and action sequences. Quality and linguistic precision were emphasized to exceed benchmark QA requirements. • Annotated 2,000+ samples for LLM fine-tuning across robotic and video game domains. • Applied reasoning to capture complex temporal and action relationships. • Supported embodied AI models through high-accuracy annotation. • Consistently achieved a 98% QA accuracy rate.

2025 - 2025

Media Post Description Annotator

LabelboxTextClassification

In this media annotation project, I reviewed over 5,000 AI-generated descriptions of user-posted media. Using a decision tree framework, I checked text labels for accuracy, user interest alignment, and salient content. My work led to measurable improvements in content recommendation precision. • Classified descriptive text by topic, theme, and user interest area. • Achieved 96% annotation accuracy, supporting improved model output. • Detected hallucinations and low-salience details in AI descriptions. • Generated significant uplift in content recommendation precision.

In this media annotation project, I reviewed over 5,000 AI-generated descriptions of user-posted media. Using a decision tree framework, I checked text labels for accuracy, user interest alignment, and salient content. My work led to measurable improvements in content recommendation precision. • Classified descriptive text by topic, theme, and user interest area. • Achieved 96% annotation accuracy, supporting improved model output. • Detected hallucinations and low-salience details in AI descriptions. • Generated significant uplift in content recommendation precision.

2024 - 2025

Reels Plugin Annotation (Social Media Data)

Video

I participated in the Reels Plugin project, annotating over 5,000 social media video links and AI-generated responses. Assignments required relevance assessment, alignment to user prompts, and application of nuanced video labeling criteria. My work directly contributed to increasing video recommendation accuracy and reducing irrelevant content. • Assessed video and response labels for helpfulness and prompt alignment. • Applied content, caption, and subtitle standards for video annotation. • Achieved 98% annotation accuracy and reduced irrelevant recommendations. • Ensured strict compliance with project and platform guidelines.

I participated in the Reels Plugin project, annotating over 5,000 social media video links and AI-generated responses. Assignments required relevance assessment, alignment to user prompts, and application of nuanced video labeling criteria. My work directly contributed to increasing video recommendation accuracy and reducing irrelevant content. • Assessed video and response labels for helpfulness and prompt alignment. • Applied content, caption, and subtitle standards for video annotation. • Achieved 98% annotation accuracy and reduced irrelevant recommendations. • Ensured strict compliance with project and platform guidelines.

2024 - 2024

Trigger Plug-in Annotator (LLM Training)

Text

During the Trigger Plug-in Annotation project, I reviewed and rated over 8,000 chatbot responses for LLM training. I assessed each response for helpfulness, honesty, and harmlessness, and ranked them for model evaluation. Search summarization plugins were also reviewed for factual accuracy. • Evaluated large-scale chatbot data for prompt/response quality. • Ensured responses met 3H standards: honesty, harmlessness, helpfulness. • Supported LLM evaluation and RLHF via ranked annotations. • Maintained a project annotation accuracy of 93%.

During the Trigger Plug-in Annotation project, I reviewed and rated over 8,000 chatbot responses for LLM training. I assessed each response for helpfulness, honesty, and harmlessness, and ranked them for model evaluation. Search summarization plugins were also reviewed for factual accuracy. • Evaluated large-scale chatbot data for prompt/response quality. • Ensured responses met 3H standards: honesty, harmlessness, helpfulness. • Supported LLM evaluation and RLHF via ranked annotations. • Maintained a project annotation accuracy of 93%.

2024 - 2024

Education

Work History