Ezeatu E.

Key Skills

Software

Scale AI

Labelbox

Label Studio

Roboflow

Argilla

OneForma

Mercor

CVAT

Don't disclose

Top Subject Matter

LLM response evaluation

Rlhf Domain Expertise

QA and hallucination detection

Top Data Types

Video

Image

Text

Document

Top Task Types

Segmentation

Data Collection

Evaluation/Rating

RLHF

Bounding Box

Entity (NER) Classification

Freelancer Overview

AI Response Evaluator (Human Trainer) at Confidential AI Research Company (via Mercor), Remote. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Scale AI, CVAT, and Don't disclose. Education includes Bachelor of Science, Olabisi Onabanjo University (2023). AI-training focus includes data types such as Text and Document and labeling workflows including Evaluation, Rating, and RLHF.

Labeling Experience

AI Response Evaluator (Human Trainer)

Scale AIText

Evaluated AI-generated responses using a human-in-the-loop RLHF style rubric and assigned quality ratings on a 1–7 scale. Assessed alignment with desired model behavior across diverse conversation types, focusing on tone, helpfulness, honesty, writing style, and instruction-following accuracy. Produced structured, segment-level feedback by identifying both aligned and misaligned parts and providing written rationales for each label. • Rated response quality on a 1–7 quality scale. • Applied rubrics covering tone, helpfulness, honesty, style, and instruction adherence. • Wrote well-reasoned rationales highlighting aligned/misaligned segments. • Composed 4–5 sentence user-intent summaries per task to reflect conversational context.

2026 - Present

AI training

VideoSegmentation

Performed dense and coarse video annotation including temporal action segmentation, split/merge boundary enforcement, and multi-label classification

2025 - 2026

AI Data Annotation Specialist

Don't discloseDocumentRLHFEntity (NER) Classification

Applied RLHF methodologies to evaluate text, image, and multimodal datasets with a focus on instruction following, truthfulness, and response quality. Flagged ambiguous annotation edge cases and collaborated with QA teams to refine guidelines and maintain quality benchmarks. Supported pose estimation and video/media labeling workflows alongside computer vision segmentation tasks.• Engineered 10,000+ bounding box and polygon segmentations using CVAT and Labelbox.• Flagged 200+ ambiguous edge cases and exceeded 95% quality benchmarks through guideline refinement.• Processed 5,000+ keypoint and landmark annotations for human pose estimation training.• Transcribed and timestamped audio/video content while labeling actions and events across video frames.

2025 - 2025

Education

Work History