Ash Chapman

Key Skills

Software

Data Annotation Tech

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Task Types

Text Generation

Text Summarization

Freelancer Overview

I've contributed to multiple projects aimed at producing ideal LLM responses by grading several existing responses against specific, atomic rubrics. I have experience both in analyzing responses against existing rubrics and in creating rubrics myself. I also have contributed to video-training for various AI models, and I've worked on labelling conversations between users and LLMs according to the correct company-specified categories.

Labeling Experience

Ranking LLM Image Comprehension

Data Annotation TechImageObject DetectionEvaluation Rating

I compared different LLM responses to prompts requiring the LLMs to comprehend an attached image. This included ranking the LLMs based on accurate descriptions of the images, accurate understanding of what the images represented, and accurate implementation of that knowledge into their responses.

I compared different LLM responses to prompts requiring the LLMs to comprehend an attached image. This included ranking the LLMs based on accurate descriptions of the images, accurate understanding of what the images represented, and accurate implementation of that knowledge into their responses.

2025

Testing LLM capability with complex trivia

Data Annotation TechTextText SummarizationEvaluation Rating

I have worked with several projects intended to find the weak points of existing LLMs by creating precise, complex trivia questions with only one correct answer that an LLM will get correct only 1 or 2 times out of 4 attempts. This requires fine-tuning the prompt, multiple iterations of each trivia question, and identifying where LLMs tend to get tripped up and why.

I have worked with several projects intended to find the weak points of existing LLMs by creating precise, complex trivia questions with only one correct answer that an LLM will get correct only 1 or 2 times out of 4 attempts. This requires fine-tuning the prompt, multiple iterations of each trivia question, and identifying where LLMs tend to get tripped up and why.

2025

Categorizing LLM conversations with users

Data Annotation TechTextClassificationEmotion Recognition

I participated in multiple projects with the intention to categorize conversations between LLMs and users into pre-specified categories. This involved reading complex, detailed instructions, and accurate identifying the emotion and intent behind user queries in order to categorize the conversations accurately.

I participated in multiple projects with the intention to categorize conversations between LLMs and users into pre-specified categories. This involved reading complex, detailed instructions, and accurate identifying the emotion and intent behind user queries in order to categorize the conversations accurately.

2025

Education

Work History