Lucky N.

Key Skills

Software

OneForma

Appen

Other

Top Subject Matter

AI/ML image datasets (vehicles, products/graphics, animals)

LLM output evaluation and safety/risk assessment

LLM response ranking and quality evaluation

Top Data Types

Text

Image

Document

Audio

Top Task Types

Text Summarization

Classification

Evaluation/Rating

Freelancer Overview

Brings 9+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Appen, OneForma, Sapien, and others Education includes a Bachelor of Medicine & Bachelor of Surgery, University of Calabar (2024). AI-training focuses on data types such as Image, Text, and Document, and labeling workflows including Classification, Evaluation, Annotation, and Rating.

Labeling Experience

Data Annotation

OtherImageClassification

Performed image annotation and categorical labeling for AI/ML training datasets by classifying images into predefined client categories. Applied labeling guidelines by analyzing visual attributes such as background, product setting, and graphic overlays. Maintained high annotation consistency and quality across multiple batches. • Classified 500+ images into defined categories per client guidelines. • Labeled 4000+ vehicle images by angle and view with accurate tagging. • Identified animal categories from multiple-choice lists for training data. • Maintained 95%+ accuracy across annotation batches.

2025 - 2025

Project HEMC — High-volume LLM output rating with safety screening

AppenTextEvaluation/Rating

Evaluated LLM outputs across helpfulness, accuracy, coherence, and safety using remote structured rubrics. Rated responses for correctness, completeness, and instruction adherence to support quality improvements. Flagged harmful content including misinformation, stereotypes, and PII while maintaining consistency and impartiality across high-volume annotation tasks. • Assessed instruction following, correctness, and completeness under the rubric. • Used safety checks to identify misinformation, stereotypes, and privacy leaks (PII). • Maintained consistent labeling standards across large batches. • Ensured impartial scoring aligned to project guidance.

2024 - 2024

Project Artemis — 3D framework evaluation and ranking for LLM responses

OneFormaTextEvaluation/Rating

Evaluated and ranked thousands of AI-generated responses using a structured 3-dimensional framework (Harmlessness, Honesty, Helpfulness). Applied priority hierarchies to resolve conflicting quality signals and produced an absolute quality score for each response. Identified incomplete answers, grammar errors, factual inaccuracies, and harmful content based on the project rubric. • Scored and ranked outputs using Harmlessness/Honesty/Helpfulness dimensions. • Used priority ordering to handle conflicts between quality signals. • Flagged factual errors, harmful content, and incompleteness. • Maintained consistent application of the structured scoring methodology.

2024 - 2024

Project Aralia — Harmlessness/Honesty/Helpfulness ranking and 1–7 rating

AppenTextEvaluation/Rating

Conducted side-by-side comparisons of AI-generated responses and ranked outputs across the Harmlessness, Honesty, and Helpfulness dimensions. Assigned quality ratings on a 1–7 scale and verified factual accuracy using independent research when needed. Flagged outputs containing bias, stereotypes, sensitive advice, or content likely to cause harm. • Used comparative ranking to determine the better response among multiple outputs. • Applied a 1–7 quality rating scale per project instructions. • Performed independent fact checks to validate truthfulness. • Flagged unsafe or potentially harmful content categories for review.

2024 - 2024

Project Hermes — LLM output rating across multiple quality/safety dimensions

OtherTextEvaluation/Rating

Rated thousands of LLM outputs across a wide range of prompts using a rubric-based scoring approach. Assessed helpfulness, harmlessness, hallucination risk, completeness, succinctness, logical cohesion, and truthfulness according to project guidelines. Applied consistent evaluation criteria to support model iteration. • Evaluated responses for hallucinations, completeness, and adherence to quality rubrics. • Scored truthfulness and logical coherence across many prompt types. • Assessed safety and harmlessness to flag problematic outputs. • Used structured project guidelines to maintain consistent scoring at scale.

2023 - 2024

Show more

Education

Work History