Mingdong P.

Key Skills

Software

Other

Top Subject Matter

Operations & Supply Chain / Maritime Logistics

Business Administration & Analytics

Regulatory Compliance & Risk Management

Top Data Types

Text

Top Task Types

RLHF

Data Collection

Evaluation/Rating

Red Teaming

Text Generation

Object Detection

Question Answering

Bounding Box

Text Summarization

Freelancer Overview

I am an incoming AI Trainer and Data Evaluator, certified by Google (AI Professional & Essentials) and trained in business analytics through Wharton’s AI for Business specialization. My background is unconventional: I spent years as a licensed ship officer on international routes, managing safety and navigating under strict global regulations. This career built a lifelong habit of following complex checklists exactly, working independently for long stretches, and catching small errors before they lead to operational failures. I am now applying this exact same zero-variance discipline to auditing model outputs, testing prompts, and tracking hallucinations. Though transitioning into this space, I bring hands-on experience in structured prompt testing, blind A/B model comparisons, and technical fact-checking. I specialize in identifying subtle reasoning errors, logic gaps, and misattributions in model responses. Backed by my maritime technical background and full professional English proficiency, I excel at writing clear, objective, and evidence-based justifications to explain precisely where a model failed or followed instructions. I have a quiet, dedicated remote setup and am looking to leverage my rigorous operational discipline for high-quality data labeling projects.

Labeling Experience

Multi-Model Blind A/B Testing & Logical Error Auditing

TextEvaluation/Rating

Conducted an independent, structured evaluation project focused on stress-testing the logical boundaries and constraint adherence of frontier LLMs (including Gemini and Claude). Designed and executed a series of 50+ bilingual (English/Chinese) prompts using structured prompting techniques like Chain-of-Thought (CoT) and negative constraints to evaluate multi-step reasoning capabilities. My hands-on work involved running blind A/B preference testing, meticulously cataloging model outputs based on logical consistency, information density, and strict adherence to formatting instructions. I maintained a granular tracking log to analyze and categorize text hallucinations, specifically capturing source misattributions and logical overgeneralizations in technical summaries. For every evaluation, I drafted objective, evidence-based justifications in professional English, breaking down the exact mathematical or logical failures of the underperforming model. This project demonstrated my ability to apply zero-variance operational discipline—honed during my career managing complex maritime data logs—to high-density AI data auditing.

2026 - 2026

Education

Work History