Joshua Grant

Key Skills

Software

Data Annotation Tech

Top Subject Matter

No subject matter listed

Top Data Types

Text

Video

Image

Top Label Types

Text Generation

Evaluation Rating

Question Answering

Text Summarization

Freelancer Overview

I am an analytical professional with hands-on experience in AI data annotation, evaluation, and quality assurance. My current freelance work focuses on designing fine-grained evaluation criteria, crafting and testing challenging prompts, and conducting meticulous fact-checking of AI model outputs across factual, creative, and reasoning domains. I excel at reviewing and annotating nuanced content for accuracy, clarity, safety, and alignment with system instructions, drawing on my strong background in regulated environments such as rail operations, aviation security, and military service. With a foundation in clear written communication, risk identification, and evidence-based review, I bring a detail-oriented and process-driven approach to every project. I am comfortable working independently, adapting to new workflows, and continuously improving quality standards in the fast-evolving field of AI training data. I am currently studying a Level 4 Paralegal Qualification and Python. I study classical Latin, and I have been for nearly 4 years now. It is my aim to enter into AI Red Teaming and AI security as I progress in my career in this sector.

Labeling Experience

Professional API Prompt

Data Annotation TechTextText GenerationEvaluation Rating

For this project, I designed professional domain API prompts, including system instructions, input data and user input to test two models to surface an instruction following failure. The system instructions needed a real world application within the specific domain and to be difficult enough to make the models struggle.

For this project, I designed professional domain API prompts, including system instructions, input data and user input to test two models to surface an instruction following failure. The system instructions needed a real world application within the specific domain and to be difficult enough to make the models struggle.

2025 - 2025

<= 20k Token Long Instruction Following

Data Annotation TechTextQuestion AnsweringText Generation

This project involves inputting a prompt with 2+ instructions per turn, with the first turn including a document of some kind, either creative, professional or something else, then prompting questions based on that specific text across multiple turns, up to 10 turns, with the intention to surface a major failure along one of the axes, with a minimum of one major instruction failure per task, then editing the preferred response.

This project involves inputting a prompt with 2+ instructions per turn, with the first turn including a document of some kind, either creative, professional or something else, then prompting questions based on that specific text across multiple turns, up to 10 turns, with the intention to surface a major failure along one of the axes, with a minimum of one major instruction failure per task, then editing the preferred response.

2025

Fact Checking

Data Annotation TechTextText GenerationEvaluation Rating

The project involves creating a breakdown of response model claims and verifying them against authoritative sources, then labelling them against correct, incorrect, disputed and unsupported criteria. This involves juxtaposing the two models, choosing a the 'better' one across multiple axes and giving comprehensive feedback.

The project involves creating a breakdown of response model claims and verifying them against authoritative sources, then labelling them against correct, incorrect, disputed and unsupported criteria. This involves juxtaposing the two models, choosing a the 'better' one across multiple axes and giving comprehensive feedback.

2025

Image Based Information Q&A

Data Annotation TechImageQuestion AnsweringText Generation

This project involved uploading an open source image that contained a particular data type, from concrete to abstract, then prompting with a question about that particular object, resulting in two responses, from which I would rate the 'better' model based on a variety of specific criteria.

This project involved uploading an open source image that contained a particular data type, from concrete to abstract, then prompting with a question about that particular object, resulting in two responses, from which I would rate the 'better' model based on a variety of specific criteria.

2025 - 2025

Short Video Analysis

Data Annotation TechVideoQuestion Answering

This project involved rating an AI generated clip based on a provided video and user prompt to rate the model's performance against a variety of axes

This project involved rating an AI generated clip based on a provided video and user prompt to rate the model's performance against a variety of axes

2025 - 2025

Education

Work History