Ahmed W.

Key Skills

Software

Other

OpenCV AI Kit (OAK)

Top Subject Matter

AI evaluation

RLHF-style safety/compliance review

and trajectory QA for LLM systems

Top Data Types

Text

Image

Document

Top Task Types

Bounding Box

Prompt + Response Writing (SFT)

Freelancer Overview

I specialize in high-level data labeling, model alignment, and human-in-the-loop quality assurance. My experience spans complex multi-pass evaluation workflows, including Reinforcement Learning from Human Feedback (RLHF), side-by-side prompt ranking, and auditing long-form AI reasoning trajectories. I am deeply adept at executing rigid evaluation guidelines to detect subtle data anomalies such as factual deviations, logical loops, and output truncations, scoring model performance against strict enterprise rubrics to ensure data integrity and safety compliance.

Labeling Experience

Data Quality Auditor & AI Evaluation Reviewer (Independent Contractor / Remote)

OtherText

Performed multi-pass human-in-the-loop evaluation of complex multi-turn AI reasoning trajectories to detect logic loops, factual deviations, and output truncation across long step sequences. Scored model behaviors using enterprise rubrics covering factual correctness, algorithmic efficiency, and logical consistency to support quality assurance and iteration. Verified grounding and safety compliance by auditing context data pipelines to prevent future-dated fabrication and guideline leakage while flagging hallucinated syntax and model overreach.• Evaluated 20+ step trajectories for reasoning quality and stability• Applied rubric-based scoring and classification to annotate model behavior• Conducted compliance checks for policy, safety, and grounding adherence• Identified and reported model logic errors for remediation

2025 - Present

AI Data Curator & NLP Analyst (Academic & Independent Projects)

OpenCV AI Kit (OAK)ImageBounding Box

Built and optimized a computer vision dataset for a road damage detection system containing 400+ assets. Applied programmatic image preprocessing filters using OpenCV (e.g., Gaussian blurring and edge detection) to isolate inconsistencies and improve annotation quality before downstream fine-tuning. Manually audited bounding boxes to ensure consistency across the dataset and reduce label noise for reliable model training.• Produced and validated bounding box labels for road damage assets• Used OpenCV-based preprocessing to surface annotation inconsistencies• Performed quality checks to ensure label consistency prior to fine-tuning• Curated dataset readiness for computer vision model training workflows

2024 - 2025

Education

Work History