Ayoub Tanghort

Key Skills

Software

Appen

Clickworker

Mindrift

Telus

Internal/Proprietary Tooling

Scale AI

OneForma

Top Subject Matter

No subject matter listed

Top Data Types

Document

Medical Dicom

Text

Top Task Types

Classification

Emotion Recognition

Evaluation Rating

RLHF

Translation Localization

Freelancer Overview

I’m a multilingual AI-training specialist with 4 years of experience in data labeling, LLM evaluation, and localization QA. Fluent in Arabic, English, German, French, and Russian, I create and audit high-fidelity datasets that power next-gen language and speech models. Highlights include reducing hallucination rates from 22 % → 6 % on an Arabic LLM by redesigning evaluation rubrics and prompt-grounding workflows, and producing 200 + studio-quality German speech prompts now used in a TTS model with < 2 % word-error rate. Skilled with Labelbox, Hugging Face Datasets, Python/pandas, and live Power BI dashboards, I combine linguistic precision with data-driven insights to deliver accurate, culturally nuanced training data on tight timelines. Certified Google Cloud Data Engineer and Salesforce Administrator, I thrive in cross-functional teams where measurable quality and rapid iteration drive AI performance.

Labeling Experience

Dr Strange Annotation Project for Native German Speakers

OneformaMedical DicomDiagnosis

Contribute​ to the advancement​ of​ AI​ in Healthcare​ by annotating clinical data, transcriptions, and dialogues​ to enhance AI-driven medical documentation.​

Contribute​ to the advancement​ of​ AI​ in Healthcare​ by annotating clinical data, transcriptions, and dialogues​ to enhance AI-driven medical documentation.​

2025

Citadel Choir - DE-de

Scale AITextPrompt Response Writing SFT

writing a prompt in German, evaluate 2 responses from two different model, and than providing a real human answer.

writing a prompt in German, evaluate 2 responses from two different model, and than providing a real human answer.

2025

Starmie

Scale AIMedical DicomRLHFPrompt Response Writing SFT

Exclusive community of over 1800 experts—from accomplished PhDs to brilliant researchers and top-performing undergraduates—all dedicated to challenging the limits of AI with expert-level questions in Humanity’s

Exclusive community of over 1800 experts—from accomplished PhDs to brilliant researchers and top-performing undergraduates—all dedicated to challenging the limits of AI with expert-level questions in Humanity’s

2025

BABEL Multilingual LLM Evaluation & RLHF (Arabic / German)

AppenTextRLHFEvaluation Rating

Evaluated 5000+ model outputs in Arabic and German for factual accuracy, relevance, brevity, and stylistic correctness. Redesigned the evaluation rubric and prompt-grounding workflow, which reduced hallucination errors from 22 % to 6 %. Created gold-standard answer sets for RLHF fine-tuning and ran cross-rater calibration sessions to raise inter-annotator agreement to 92 %.

Evaluated 5000+ model outputs in Arabic and German for factual accuracy, relevance, brevity, and stylistic correctness. Redesigned the evaluation rubric and prompt-grounding workflow, which reduced hallucination errors from 22 % to 6 %. Created gold-standard answer sets for RLHF fine-tuning and ran cross-rater calibration sessions to raise inter-annotator agreement to 92 %.

2024

Education

Work History