Ule M.

Key Skills

Software

AWS SageMaker

Google Cloud Vertex AI

Mercor

Micro1

VoTT

Top Subject Matter

Healthcare

Software

E-Commerce - Product Categorization & Customer Support

Top Data Types

Image

Text

Computer Code Programming

Top Task Types

RLHF

Text Generation

Computer Programming/Coding

Function Calling

Fine-tuning

Freelancer Overview

Over the past year, I have developed a solid foundation in the AI lifecycle by specializing in high-quality data labeling and AI training data generation. My experience spans a variety of annotation tasks, including image segmentation, text classification, and entity tagging, all aimed at optimizing machine learning models for computer vision and natural language processing (NLP). I am adept at interpreting complex annotation guidelines quickly and maintaining a high level of precision and consistency, ensuring that the datasets delivered are of the highest standard for model training. What sets me apart is my keen eye for detail and my ability to adapt to evolving project requirements in fast-paced environments. Beyond just labeling, I have experience performing quality assurance (QA) checks to identify and rectify edge cases, significantly reducing error rates in final datasets. I am proficient with industry-standard annotation tools and possess a strong understanding of how data quality directly impacts AI performance, making me a reliable and proactive contributor to any data operations team.

Labeling Experience

AI Text Quality Assurance (QA) Specialist

TextRLHF

In this role, you act as the final human checkpoint in the training pipeline for a next-generation conversational AI model. You review complex text outputs where the AI was instructed to summarize multi-page financial documents, generate legal summaries, or write instructional guides under strict formatting constraints (e.g., "Write a 3-paragraph summary using passive voice, with no more than two adjectives per sentence"). You cross-examine the AI’s text against source documents to identify and flag subtle fabrications ("hallucinations") or logical fallacies. Rather than executing bulk data entry, you apply your knowledge of model flaws to score the text on a granular rubric and edit the winning response into a "perfect" target text. Your evaluations serve as the gold-standard data used for the critical final alignment phase before the model is deployed to production.

2025 - 2026

Education

Work History