For employers

Hire this AI Trainer

Kinza

Agency

Lead Image & Video Annotation (Workforce Management) for high-volume computer vision training data

Lahore, Pakistan

msheesr.github.io/ozlo-ai-agency/

Key Skills

Software

Labelbox

CVAT

Roboflow

Tagtog

Data Annotation Tech

Other

Label Studio

Top Subject Matter

Computer Vision & Multi-domain Image/Video Annotation

Computer Vision & Instance Segmentation

Computer Vision & Keypoint Pose Estimation

Top Data Types

Image

Text

Computer Code Programming

Top Task Types

RLHF

Evaluation/Rating

Prompt + Response Writing (SFT)

Data Collection

Computer Programming/Coding

Bounding Box

Segmentation

Point/Key Point

Red Teaming

Polygon

Classification

Tracking

Entity (NER) Classification

Cuboid

Transcription

Company Overview

Lead Image & Video Annotation (Workforce Management) for high-volume computer vision training data. Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include CVAT, Labelbox, and Label Studio. AI-training focus includes data types such as Image, Medical, and DICOM and labeling workflows including Bounding Box, Segmentation, and Point.

Security

Security Overview

we prioritize data integrity and security in all AI development and annotation workflows. Our security posture includes: Secure Work Environment: All team members utilize encrypted hardware and secure, password-protected development environments. Data Privacy: We adhere to strict internal protocols to ensure client data confidentiality. We never store, share, or expose raw data outside of the secure platforms designated by our clients. Access Control. All sensitive project access is restricted to authorized team members using secure authentication methods. Professional Standards We operate with professional transparency, ensuring all project outputs meet the high standard security requirements expected by our AI industry partners.

Labeling Experience

AI Orchestration & Multi Model Integration

Other

Computer Code Programming

Evaluation/Rating

Developed the Aetheros orchestration platform, integrating diverse proprietary and open-source models (including Gemini, DeepSeek, and Kimi) into a unified architecture. Scope: Engineered a standardized abstraction layer to manage model interoperability, context passing, and real-time task routing. Tasks Conducted cross-model performance bench marking and developed evaluation frameworks to ensure output accuracy latency optimization and adherence to system instructions. Quality Measures Implemented automated guardrails for safety and reliability, ensuring seamless data flow across heterogeneous AI environments

2026 - Present

AI Model Evaluation & RLHF Specialist

Other

Computer Code Programming

RLHF

Data Collection

Developed and validated AI-driven decision-making systems for the Order Guard Profit & Fraud Shopify application through intensive data annotation and logic training. Data Annotation Tasks Created and refined high accuracy training datasets by labeling thousands of transaction logs and merchant order patterns to optimize fraud detection logic. Model & Logic Evaluation. Performed rigorous evaluation of algorithmic responses and code logic to ensure the application accurately adhered to complex fulfillment and profit analytics (CM1/CM2) parameters. Quality Measures Enforced strict data cleaning, labeling consistency checks, In addition to data annotation performed code annotation and logical debugging to optimize the app internal algorithm and ensure high accuracy performance and stress testing protocols to maintain real time accuracy and performance stability within the Shopify ecosystem

2026 - 2026

AI Augmented Visual & Media Structuring

Roboflow

Image

Segmentation

Classification

Produced and structured high-fidelity visual assets, including 3D animations and VFX, to support training datasets for computer vision models. Our process involved meticulous frame-by-frame annotation, metadata structuring, and quality assurance to ensure visual data accuracy for photorealistic rendering and AI engagement training

2025 - 2026

Adversarial Red-Teaming for guardrail stress tests

Text

Red Teaming

Conducted adversarial red-teaming by stress-testing model guardrails with jailbreak and bias prompt sequences. Logged and evaluated weaknesses in model behavior to improve safety-related alignment. Ensured systematic coverage of adversarial prompt patterns for robust guardrail performance. • Tested jailbreak prompt sequences against guardrail behavior • Evaluated bias-related prompting and failure modes • Produced curated red-teaming insights for alignment improvements • Supported iterative safety tuning through repeated scenario runs

2024 - 2024

Pairwise Preference Curation and Likert Quality Assessments

Text

RLHF

Performed pairwise preference curation using human critiques to create ranking signals for reward-model training. Produced Likert-style quality assessments to score responses across helpfulness, accuracy, and grounding. Used these evaluations to support RLHF-style model alignment workflows. • Created pairwise comparisons between model responses • Trained reward models from curated human preference data • Scored responses using helpfulness/accuracy/grounding Likert ratings • Supported auditing and evaluation processes for alignment readiness

2024 - 2024