Andrew Mwaniki

Key Skills

Software

AWS SageMaker

Anno-Mage

Appen

Clickworker

Data Annotation Tech

Deep Systems

Humanatic

Img Lab

Labelbox

OneForma

Prodigy

CVAT

SuperAnnotate

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Geospatial Tiled Imagery

Video

Top Task Types

Action Recognition

Classification

Data Collection

Entity Ner Classification

Evaluation Rating

Freelancer Overview

With 4 years of hands-on experience in AI training data, I specialize in data labeling, annotation, and LLM evaluation for high-impact machine learning applications. I have worked across diverse domains—including NLP, computer vision, and multilingual datasets—ensuring data quality and model readiness through meticulous annotation standards. My expertise spans tools like Labelbox, Prodigy, SuperAnnotate, and Scale AI, as well as programming with Python for automation and QA in annotation workflows. I’ve led teams, trained annotators, and developed guidelines tailored to domain-specific AI systems. My standout contributions include curating multilingual sentiment datasets, designing prompt evaluations for LLMs like GPT-4 and Claude, and building automation tools for classification QA. I bring a deep understanding of prompt engineering, bias mitigation, and model alignment testing, which positions me as a valuable contributor to teams focused on creating accurate, ethical, and scalable AI systems.

Labeling Experience

LLM Prompt Evaluation and Toxicity Annotation (Meta AI)

SuperannotateTextTrackingTranslation Localization

Contracted through a global vendor to support Meta’s LLaMA model training pipeline. Evaluated model responses to user prompts and labeled completions for safety, toxicity, factual consistency, and helpfulness. Used structured rubrics to rank outputs and flag problematic completions. Played a key role in refining model alignment with human values and ethical AI guidelines.

Contracted through a global vendor to support Meta’s LLaMA model training pipeline. Evaluated model responses to user prompts and labeled completions for safety, toxicity, factual consistency, and helpfulness. Used structured rubrics to rank outputs and flag problematic completions. Played a key role in refining model alignment with human values and ethical AI guidelines.

2024 - 2024

Multilingual Sentiment Classification

LabelboxTextClassificationQuestion Answering

Worked on a sentiment classification project for a Google AI-backed initiative targeting underrepresented African languages. Annotated over 10,000 social media posts in Swahili, English, and French for sentiment and intent. Developed custom sentiment taxonomy for local dialects and slang. Played a lead role in ensuring linguistic consistency and cultural context accuracy across regional datasets.

Worked on a sentiment classification project for a Google AI-backed initiative targeting underrepresented African languages. Annotated over 10,000 social media posts in Swahili, English, and French for sentiment and intent. Developed custom sentiment taxonomy for local dialects and slang. Played a lead role in ensuring linguistic consistency and cultural context accuracy across regional datasets.

2023 - 2023

Product Image Tagging for E-Commerce Visual Search

CVATImageBounding BoxEntity Ner Classification

Collaborated with an Amazon AI subcontractor to annotate over 15,000 product images for use in training computer vision algorithms powering visual search. Tasks included drawing bounding boxes around products, tagging features such as color, texture, and brand, and ensuring alignment with Amazon’s taxonomy guidelines.

Collaborated with an Amazon AI subcontractor to annotate over 15,000 product images for use in training computer vision algorithms powering visual search. Tasks included drawing bounding boxes around products, tagging features such as color, texture, and brand, and ensuring alignment with Amazon’s taxonomy guidelines.

2022 - 2023

Education

Work History