a
aipro_trainer

PROMPTandTRAIN

@aipro_trainer

LLMTrainer, RLHFSpecialist, AIAnnotatorPro

Frankreich
Französisch, Englisch
Einige Informationen werden in englischer Sprache angezeigt.
Über mich
my name is Antoine ,I've spent the last year and a half evaluating AI responses for Outlier AI and Alignerr — over 5,000 of them, holding myself to a 95% accuracy bar the whole time. Along the way I started building adversarial prompts to stress-test models, and found 40+ recurring failure patterns that ended up feeding into real retraining decisions. I also annotate NLP data and fact-check across science, law, finance and tech. English fluent, Yoruba native. I care about getting the small details right, not just hitting a quota.... Mehr lesen

Kompetenzen

a
aipro_trainer
PROMPTandTRAIN
offline • 

Meine Dienstleistungen

KI-Technologie-Beratung
I will train evaluate your ai model responses with rlhf and prompt engineering
Ab10 $ / ProjektWeitere Details

Arbeitserfahrung

Micro_Image Technologies

AI Data Annotator & Content Specialist

Micro Image Technologies • Freiberufler

Aug 2024 - Present • 1 yr 11 mos

Annotate 5,000+ data points for NLP & ML pipelines — maintaining 95%+ consistency across 15+ simultaneous projects (scaled from 2 to 15 projects while keeping quality flat).  Review and correct AI-generated content — improved output quality baseline from 82% to 95% through refined annotation standards.  Conduct multi-source fact-checking in specialized domains (science, law, finance, tech) — validated 2,000+ claims with 100% source documentation.  Manage timeline pressures without sacrificing quality — averaged 5-day turnaround on urgent projects with zero quality lapses or deadline misses. KEY ACHIEVEMENTS  Achieved and maintained 95% accuracy across 5,000+ evaluations (improved baseline by 7 percentage points through methodology refinement)  Mastered 10+ specialized evaluation domains from basics to expert level in under 12 months  Scaled from 1 to 2 concurrent platforms while increasing total throughput 40% with zero quality degradation  Identified 40+ systematic model failure modes that directly informed 3+ released model improvements  Zero missed deadlines and zero quality rejections across 18+ months of continuous work  Built expertise spanning RLHF workflows, NLP annotation, conversational AI testing, and factchecking verification