Paul M.

Key Skills

Software

CloudFactory

Label Studio

Top Subject Matter

Health care and medicine

Education, academics, STEM

Top Data Types

Text

Medical Dicom

Document

Top Task Types

Entity (NER) Classification

RLHF

Fine-tuning

Freelancer Overview

PROFESSIONAL PROFILE Detail-oriented AI training contributor and data annotation specialist with a strong foundation in analytical reasoning, structured classification, and high-accuracy quality control — developed through rigorous medical studies at the University of Nairobi. Experienced in producing and evaluating precisely structured content against defined guidelines, flagging inconsistencies, and maintaining annotation quality across large datasets. Brings domain expertise in healthcare and biomedical data, strong written English (C1/C2), and a disciplined, methodical work ethic that translates directly into reliable, high-throughput labelling output. CORE COMPETENCIES ◆ Text & NLP Annotation ◆ Medical / Clinical Data Labelling ◆ Quality Assurance & Accuracy Review ◆ Instruction Following ◆ RLHF / Preference Ranking ◆ Fact-Checking ◆ Content Classification ◆ Image & Document Annotation ◆ Attention to Detail ◆ Remote Workflow Management DATA LABELLING & ANNOTATION EXPERIENCE Freelance AI Data Contributor 2023 – Present Independent — Remote Annotated text datasets for NLP tasks including intent classification, entity recognition, and sentiment labelling, maintaining >97% inter-annotator agreement Reviewed and ranked AI-generated responses for quality

Labeling Experience

AI Response Evaluation & Preference Ranking — RLHF Feedback Dataset

TextEntity (NER) Classification

Freelance contributor to Reinforcement Learning from Human Feedback (RLHF) projects, providing human preference signals used to fine-tune large language models. Core tasks involve reviewing pairs or sets of AI-generated responses to a given prompt and ranking them based on defined quality criteria — including factual accuracy, logical coherence, instruction-following, appropriate tone, and absence of hallucinated content. Evaluated approximately 300–500 response pairs per active month across topics spanning general knowledge, medical information, educational content, and creative writing. Applied structured rubrics provided per project, escalating edge cases and ambiguous prompts to QA leads for calibration. Maintained consistent ranking alignment with team benchmarks throughout, demonstrating strong evaluative judgment and an understanding of what constitutes high-quality model output versus subtly flawed responses.

2025 - Present

NLP Tasks datasets annotation

TextEntity (NER) Classification

Ongoing freelance engagement performing Named Entity Recognition (NER) annotation and text classification across medical and general-knowledge datasets used to train large language models. Tasks include identifying and tagging clinical entities (diseases, symptoms, medications, anatomical terms, procedures) within unstructured text, as well as classifying intent, sentiment, and semantic category for general NLP corpora. Project scope spans approximately 500–800 annotated text samples per active month, covering clinical case summaries, biomedical literature excerpts, patient-facing health content, and open-domain Q&A pairs. Quality measures include strict adherence to annotation rubrics supplied per project, self-review passes before submission, and regular calibration against gold-standard labels to maintain inter-annotator agreement above 96%. Feedback from QA reviewers is actively incorporated to continuously improve label consistency and edge-case handling.

2025 - Present

Educational Content Classification & Quality Review

TextEntity (NER) Classification

During tenures as a Peer Teacher at Kerugoya Boys High School, developed and systematically classified large volumes of academic text content — including questions, explanations, worked examples, and student responses — across multiple subject disciplines. Tasks closely mirror professional data annotation workflows: categorising content by subject, difficulty level, reasoning type (factual recall, applied reasoning, critical analysis), and quality tier. Reviewed and assessed student-written responses against defined marking criteria, identifying errors, inconsistencies, and missing information — a process directly analogous to AI output quality review and annotation. Produced structured written feedback aligned to per-subject rubrics, maintaining consistency across hundreds of assessments per engagement period. Developed sensitivity to subtle differences in response quality, tone, and accuracy that is directly applicable to preference labelling and classification tasks.

2025 - 2026

Education

Work History