Stephanie Trafford

Key Skills

Software

Appen

CloudFactory

Data Annotation Tech

Google Cloud Vertex AI

Remotasks

Scale AI

Telus

V7 Labs

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Text

Video

Top Task Types

Computer Programming/Coding

Prompt + Response Writing (SFT)

Red Teaming

RLHF

Text Generation

Freelancer Overview

I have extensive experience in data labeling and AI training across a range of advanced projects involving natural language processing, multimodal input, and reinforcement learning from human feedback (RLHF). At Scale AI, I contributed to the Dolphin Genesis project, where I wrote diverse prompts and evaluated model responses—selecting the most accurate and informative replies to train alignment systems. This work was critical to improving the quality of LLM-generated output through careful judgment and comparative assessments. My experience also spans multimodal projects like “Biscuits with Rubrics,” where I developed image-based prompts and designed scoring rubrics to assess model understanding. In tasks such as Laurien Moon, I crafted prompts that intentionally challenged models, provided ground-truth full answers (GTFA), and wrote effective hints to guide model correction. Most recently, I worked on Pegasus HLE, where I created complex, high-level prompts and solutions to test model reasoning at the PhD/Master’s level—ensuring deep alignment with high-level academic standards. These diverse projects reflect my strong analytical skills, instructional clarity, and commitment to advancing safe and intelligent AI systems.

Labeling Experience

Advanced Data Annotation and RLHF for LLMs

Scale AITextText GenerationText Summarization

In the Pegasus HLE project, I created advanced PhD/Master’s-level prompts to evaluate and improve LLM reasoning in high-complexity tasks. Responsibilities included writing challenging questions, crafting ideal ground-truth answers, and providing strategic hints to guide model thinking. This work supported alignment and fine-tuning for models in academic and professional domains.

In the Pegasus HLE project, I created advanced PhD/Master’s-level prompts to evaluate and improve LLM reasoning in high-complexity tasks. Responsibilities included writing challenging questions, crafting ideal ground-truth answers, and providing strategic hints to guide model thinking. This work supported alignment and fine-tuning for models in academic and professional domains.

2025 - 2025

Education

Work History