Muhammad A.

Key Skills

Software

Data Annotation Tech

Humanatic

Scale AI

SuperAnnotate

Telus

Top Subject Matter

Egocentric video labeling for computer vision action/interaction datasets

AI training data preparation for computer vision and NLP/RLHF

Legal Services & Contract Review

Top Data Types

Video

Audio

Image

Text

Document

Top Task Types

Action Recognition

RLHF

Freelancer Overview

AI Data Annotator — Egocentric Video Labeling (Atlas Capture). Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Atlas Capture, N, and A. Education includes Secondary School Certificate, WAEC/NECO. AI-training focus includes data types such as Video and Audio and labeling workflows including Action Recognition and RLHF.

Labeling Experience

AI Training Data Contributor (Independent/Remote)

AudioRLHF

Contributed to AI training datasets by completing image and object labeling tasks for computer vision development and by evaluating AI-generated text responses. Verified audio recordings through transcription and validation to support NLP model training across multiple task categories. Assessed responses for accuracy, helpfulness, safety, and cultural appropriateness as part of RLHF workflows. • Completed image annotation, bounding box labeling, and object classification tasks per platform guidelines. • Evaluated text responses for accuracy, helpfulness, safety, and cultural appropriateness within RLHF. • Transcribed and verified audio recordings for NLP training across task categories. • Maintained high accuracy ratings across batches to access higher-tier task categories.

2025 - Present

AI Data Annotator — Egocentric Video Labeling (Atlas Capture)

VideoAction Recognition

Reviewed and labeled first-person (egocentric) video segments of humans performing physical tasks using dense and coarse, segment-level annotation rules. Calibrated start and end timestamps for each segment, ensured compliance with duration and multi-action limits, and applied quality review checks for accuracy and forbidden verb usage. Distinguish task-relevant interactions from No Action segments and maintain annotation integrity with approved fallback nouns when objects are unidentified. • Labeled goal-oriented hand-object interactions with action verbs and object descriptions using dense/coarse frameworks. • Performed timestamp correction to avoid overlap or truncation and to accurately capture interaction scope. • Conducted QA error detection including correcting grammatical issues and hallucinated actions in existing annotations. • Managed episode batches via a platform task queue while following segment rules (e.g., 20-second max duration, 2 actions/segment).

2025 - Present

Education

Work History