Precious Adeyemi

Key Skills

Software

Data Annotation Tech

Roboflow

Telus

Scale AI

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Document

Image

Text

Top Label Types

Action Recognition

Audio Recording

Classification

Data Collection

Diagnosis

Evaluation Rating

Object Detection

Polygon

Prompt Response Writing SFT

Question Answering

Segmentation

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I have a strong background in AI-driven product development and market research, with hands-on experience leading teams to build and scale technology solutions in domains such as computer vision for skin analysis and fintech platforms. My work as CEO of Beeva AI involved overseeing the creation of a skin analysis app, where I collaborated closely with cross-functional teams to ensure data quality and user-centric design, which is crucial for effective AI training data. I am skilled in user research, documentation, and user acceptance testing, and have a solid understanding of agile methodologies and product management best practices. My experience spans working with diverse datasets, aligning product features with customer needs, and driving innovation through data-driven decision making. I am passionate about leveraging data annotation and labeling to enhance AI solutions and deliver real-world impact.

Labeling Experience

Named Entity Recognition & RLHF Annotation for Healthcare LLM

LabelboxTextEntity Ner ClassificationText Summarization

Worked on fine-tuning a healthcare-focused LLM to improve medical question answering and summarization accuracy. Key contributions: - Annotated medical entities (symptoms, medications, procedures, conditions) using NER frameworks - Classified patient queries by urgency level and medical specialty - Created high-quality prompt-response pairs (SFT) for fine-tuning - Performed RLHF ranking tasks to improve response helpfulness and safety - Summarized clinical notes into structured medical reports - Conducted red teaming to identify hallucinations and unsafe medical outputs - Evaluated model outputs for factual accuracy, safety compliance, and clarity This work directly supported the deployment of a safer, domain-specific medical AI assistant.

Worked on fine-tuning a healthcare-focused LLM to improve medical question answering and summarization accuracy. Key contributions: - Annotated medical entities (symptoms, medications, procedures, conditions) using NER frameworks - Classified patient queries by urgency level and medical specialty - Created high-quality prompt-response pairs (SFT) for fine-tuning - Performed RLHF ranking tasks to improve response helpfulness and safety - Summarized clinical notes into structured medical reports - Conducted red teaming to identify hallucinations and unsafe medical outputs - Evaluated model outputs for factual accuracy, safety compliance, and clarity This work directly supported the deployment of a safer, domain-specific medical AI assistant.

2025 - 2025

Skin Image Data Labeling

RoboflowImagePolygonSegmentation

The project included curated image collection, structured data labeling, model development, and continuous validation to ensure accurate detection of skin tone, type, and common skin concerns. The dataset included multi-attribute annotations such as: - Skin tone classification (using standardized tone scales) - Skin type identification (oily, dry, combination, sensitive, normal) - Skin concern labeling (acne, hyperpigmentation, dark spots, redness, fine lines, texture) - Severity grading (mild, moderate, severe where applicable) - Structured metadata (e.g., lighting conditions, region, age range where consented) Quality Measures: - Standardized annotation guidelines and annotator training - Multi-level review and random quality audits - Dermatology-informed validation processes - Performance benchmarking across skin tone groups

The project included curated image collection, structured data labeling, model development, and continuous validation to ensure accurate detection of skin tone, type, and common skin concerns. The dataset included multi-attribute annotations such as: - Skin tone classification (using standardized tone scales) - Skin type identification (oily, dry, combination, sensitive, normal) - Skin concern labeling (acne, hyperpigmentation, dark spots, redness, fine lines, texture) - Severity grading (mild, moderate, severe where applicable) - Structured metadata (e.g., lighting conditions, region, age range where consented) Quality Measures: - Standardized annotation guidelines and annotator training - Multi-level review and random quality audits - Dermatology-informed validation processes - Performance benchmarking across skin tone groups

2023 - 2025

Multilingual Audio Transcription & Emotion Annotation for Conversational

Scale AIAudioClassificationEmotion Recognition

Project Description: Contributed to a large-scale conversational AI training project focused on improving speech recognition and emotional intelligence in AI assistants. My responsibilities included: - Transcribing over 1,500 minutes of multilingual audio (English, Nigerian Pidgin, and accented English) with 98%+ accuracy - Annotating emotional tone (e.g., neutral, frustrated, happy, confused) to improve sentiment detection models - Classifying speaker intent (complaint, inquiry, purchase intent, escalation request) - Rating AI-generated responses for clarity, empathy, and contextual correctness - Conducting quality assurance reviews on peer annotations This dataset was used to fine-tune a customer support LLM for better contextual understanding and emotionally aware responses.

Project Description: Contributed to a large-scale conversational AI training project focused on improving speech recognition and emotional intelligence in AI assistants. My responsibilities included: - Transcribing over 1,500 minutes of multilingual audio (English, Nigerian Pidgin, and accented English) with 98%+ accuracy - Annotating emotional tone (e.g., neutral, frustrated, happy, confused) to improve sentiment detection models - Classifying speaker intent (complaint, inquiry, purchase intent, escalation request) - Rating AI-generated responses for clarity, empathy, and contextual correctness - Conducting quality assurance reviews on peer annotations This dataset was used to fine-tune a customer support LLM for better contextual understanding and emotionally aware responses.

2024 - 2024

Education

Work History