Izaz Ullah

Key Skills

Software

AWS SageMaker

Appen

CrowdSource

Diffgram

Label Studio

Prodigy

Other

Don't disclose

CVAT

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Image

Text

Top Task Types

Classification

Evaluation Rating

Prompt Response Writing SFT

Question Answering

Translation Localization

Freelancer Overview

Bilingual LLM Quality Specialist with 5+ years refining AI outputs for enterprises like RWS/TrainAI. Recognized as top 5% performer for: Precision Fact-Checking: Validated 1,200+ AI responses using WHO/Reuters sources, reducing factual errors by 30% in US-based LLM projects Cultural Localization: Spearheaded Urdu-English adaptation of 500+ prompts, improving cultural appropriateness scores from 68% to 92% High-Stakes Annotation: Led a 5-person team labeling 8,000+ conversational pairs with 99.2% inter-annotator agreement Bias Mitigation: Developed Pakistan's first ethical AI guidelines for religious/financial content (adopted by 3 local startups) Core Tools: Label Studio, Prodigy, BBC Urdu Fact Check | Certified: DeepLearning.AI Prompt Engineering

Labeling Experience

Data specialist

CVATImagePolygonPoint Key Point

i worked on appen platform for different projects

i worked on appen platform for different projects

2022 - 2025

Bilingual LLM Localization for South Asian Markets

Don T DiscloseTextClassificationQuestion Answering

» Evaluated 500+ LLM responses for cultural appropriateness in Pakistani/Indian contexts » Created 50+ localization rules (e.g., Eid greetings, wedding terminologies) » Improved user satisfaction scores by 35% through tone adaptation »Done 1000+ comparison (rating/evaluation) of LLMs Dates: Dec 2023 - May 2024

» Evaluated 500+ LLM responses for cultural appropriateness in Pakistani/Indian contexts » Created 50+ localization rules (e.g., Eid greetings, wedding terminologies) » Improved user satisfaction scores by 35% through tone adaptation »Done 1000+ comparison (rating/evaluation) of LLMs Dates: Dec 2023 - May 2024

2023 - 2024

LLM Prompt Optimization (Freelance)

OtherTextDiagnosisPrompt Response Writing SFT

» Redesigned 300+ low-performing prompts across 12 clients » Achieved 15-25% accuracy improvements in specialized domains » Developed client-specific annotation guidelines for 5 LLM types Dates: Jan 2021 - Nov 2023

» Redesigned 300+ low-performing prompts across 12 clients » Achieved 15-25% accuracy improvements in specialized domains » Developed client-specific annotation guidelines for 5 LLM types Dates: Jan 2021 - Nov 2023

2021 - 2023

Ethical AI Dataset Development (Pakistani Startup)

ProdigyTextClassificationDiagnosis

» Built first Urdu sentiment dataset (10,000+ social media samples) » Identified 200+ biased phrases in Islamic finance contexts » Reduced gender bias in outputs by 40% through iterative auditing Dates: Jun 2019 - Dec 2020

» Built first Urdu sentiment dataset (10,000+ social media samples) » Identified 200+ biased phrases in Islamic finance contexts » Reduced gender bias in outputs by 40% through iterative auditing Dates: Jun 2019 - Dec 2020

2019 - 2020

Education

Work History