Samuel Agyei Agyemang

Key Skills

Software

Appen

Top Subject Matter

No subject matter listed

Top Data Types

Text

Top Task Types

Classification

Freelancer Overview

I specialize in AI training data and data annotation, with hands-on experience challenging and refining advanced language models in complex financial domains. My work at Invisible Technologies involves evaluating and improving model performance through prompt engineering, validation of factual accuracy, and detailed documentation of failure modes, particularly in financial forecasting and econometrics. I am skilled in data validation, analysis, and visualization using tools like Excel, Power BI, Tableau, SPSS, and STATA. With a strong background in finance and research, I bring a meticulous approach to ensuring high-quality, reliable datasets for AI development, and thrive in collaborative, cross-functional environments.

Labeling Experience

Data Annotation

AppenTextClassification

LLM project was fine-tuning a customer support chatbot for an e-commerce company to accurately answer product questions, process refunds, and escalate complex issues. The data labeling tasks included annotating user intent, tagging sentiment, ranking multiple AI responses for quality, rewriting ideal responses, and flagging hallucinations or policy violations. The project used approximately 120,000 high-quality annotated conversations drawn from a larger dataset of about 500,000 interactions, completed by a team of around 50 annotators and QA reviewers over three to four months. Strict quality measures were followed, including detailed annotation guidelines, annotator certification tests, inter-annotator agreement targets above 0.75, hidden gold-standard checks, and multi-layer quality audits. These controls ensured improved response accuracy, reduced hallucinations, and safe, policy-compliant model behavior.

LLM project was fine-tuning a customer support chatbot for an e-commerce company to accurately answer product questions, process refunds, and escalate complex issues. The data labeling tasks included annotating user intent, tagging sentiment, ranking multiple AI responses for quality, rewriting ideal responses, and flagging hallucinations or policy violations. The project used approximately 120,000 high-quality annotated conversations drawn from a larger dataset of about 500,000 interactions, completed by a team of around 50 annotators and QA reviewers over three to four months. Strict quality measures were followed, including detailed annotation guidelines, annotator certification tests, inter-annotator agreement targets above 0.75, hidden gold-standard checks, and multi-layer quality audits. These controls ensured improved response accuracy, reduced hallucinations, and safe, policy-compliant model behavior.

2025 - 2025

Education

Work History