Victor K.

Key Skills

Software

Scale AI

Top Subject Matter

RLHF preference modeling and conversational AI alignment

Instruction-tuned LLM data annotation and evaluation

Multilingual NMT data curation and baseline model evaluation

Top Data Types

Text

Document

Top Task Types

RLHF

Freelancer Overview

Senior AI Data Trainer, DeepMind Africa Initiative. Brings 6+ years of professional experience across legal operations, contract review, compliance, and structured analysis. Core strengths include Internal, Proprietary Tooling, and Scale AI. Education includes Bachelor of Science, University of Nairobi (2019). AI-training focus includes data types such as Text and labeling workflows including RLHF, Evaluation, and Rating.

Labeling Experience

Senior AI Data Trainer, DeepMind Africa Initiative

TextRLHF

Served as an AI Data Trainer by designing and annotating preference pairs for RLHF pipelines to improve model alignment. Ensured high-quality, consistent conversational annotations across multiple task categories and supported downstream evaluation. Collaborated with researchers on model assessment and red-teaming to surface unsafe or adversarial behaviors.• Annotated 50,000+ RLHF preference pairs with >94% inter-annotator agreement.• Created and maintained conversational AI rubrics and style guides across 8 task categories.• Flagged edge cases and hallucinations while evaluating GPT-4 and Gemini outputs.• Mentored 12 junior annotators, reducing calibration errors by 38% in 3 months.

2023 - Present

ML Data Specialist (Contract), Scale AI

Scale AIText

Completed high-complexity data annotation tasks for instruction-tuned LLM use cases including code generation, summarization, and Q&A. Maintained consistent quality through accuracy tracking and provided structured feedback on model outputs to improve coherence. Developed internal guidance for ambiguous annotation edge cases to standardize decisions across a larger annotator team.• Produced 200+ high-complexity annotations per week across code, summarization, and Q&A domains.• Maintained a personal accuracy rating of 97.2% over 18 months on the Scale Rapid platform.• Delivered structured feedback on instruction-tuned LLM responses to improve response coherence.• Authored internal guides for ambiguous edge cases used by 40+ annotators.

2021 - 2022

NLP Research Assistant, University of Nairobi

Text

Curated a Swahili-English parallel corpus to support multilingual NMT model training. Conducted baseline NLP experiments using transformer models and reported evaluation metrics to support research iteration. Assisted in academic writing for low-resource language NLP publications.• Curated a 15,000-sample Swahili-English parallel corpus for multilingual NMT training.• Ran baseline experiments with HuggingFace Transformers (BERT, mBERT) and reported metrics.• Supported research paper preparation for EMNLP and ACL workshops.• Contributed to low-resource language NLP research deliverables.

2019 - 2021

Education

Work History