Ryan Tan

Key Skills

Software

Mercor

Mindrift

Internal/Proprietary Tooling

Other

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Document

Text

Top Task Types

Bounding Box

Classification

Emotion Recognition

Evaluation Rating

Prompt Response Writing SFT

Freelancer Overview

I have three years of experience working on multimodal AI training and evaluation projects covering text, images, audio, and video. I’ve evaluated AI responses to complex scientific graphs, analyzed emotion in audio clips, reviewed multi-turn dialogues, and tested AI agents that learn to browse the web or perform simple tasks. Some projects involved checking AI-generated documents like security reports, working with LaTeX and JSON, and assessing models for safety, factual accuracy, and potential harmfulness. I’ve collaborated with companies like Mindrift, JoinStellar, DataForce (TransPerfect), and RWS, where I was often promoted to quality assurance roles. In those positions, I provided detailed feedback, helped refine labeling guidelines, and identified tool issues during pilot batches. I’m known for being precise, efficient, and adaptable, particularly on projects that require following complex instructions or applying domain expertise. I take pride in catching subtle errors that others might miss and ensuring data quality stays consistently high.

Labeling Experience

AI agent training

OtherTextBounding BoxQuestion Answering

I worked on web browsing projects (and multiple variants) for joinstellar.ai. I will keep the work intentionally vague to respect project requirements but I taught AI to browse websites and perform simple tasks like booking flights, fetching information, and adding items to cart. These projects gradually became more complex throughout the year. I was promoted to become a reviewer in most of these projects. I prioritized teaching AI skills like action and step annotation, instruction-following evaluation, reasoning assessment, webpage content understanding, safety and compliance labeling, JSON output validation, tool issue identification.

I worked on web browsing projects (and multiple variants) for joinstellar.ai. I will keep the work intentionally vague to respect project requirements but I taught AI to browse websites and perform simple tasks like booking flights, fetching information, and adding items to cart. These projects gradually became more complex throughout the year. I was promoted to become a reviewer in most of these projects. I prioritized teaching AI skills like action and step annotation, instruction-following evaluation, reasoning assessment, webpage content understanding, safety and compliance labeling, JSON output validation, tool issue identification.

2024

LLM response evaluation/generation

MindriftTextText GenerationText Summarization

I have worked on multiple projects within this time frame that involve evaluating AI responses to prompts. These projects have gradually increased in complexity throughout my time with Mindrift. It started as simple prompt/response pairs and progress to complex 10-turn dialogs within various domains. I have both crafted responses and evaluated responses based on strict rubrics. These responses ranged from simple paragraphs to complex 3-page documents. I have also served as a subject matter expert on domain specific projects like medicine, cybersecurity, and AI safety.

I have worked on multiple projects within this time frame that involve evaluating AI responses to prompts. These projects have gradually increased in complexity throughout my time with Mindrift. It started as simple prompt/response pairs and progress to complex 10-turn dialogs within various domains. I have both crafted responses and evaluated responses based on strict rubrics. These responses ranged from simple paragraphs to complex 3-page documents. I have also served as a subject matter expert on domain specific projects like medicine, cybersecurity, and AI safety.

2023

Emotion tagging/recognition

OtherAudioEmotion Recognition

I performed up to 300 emotion recognition tasks a day, which involved listening to 30-second audio clips and determining the speaker emotion, confidence, activation and identifying background noise. These clips often involved separate speakers and I had to distinguish between each speaker. I was ranked in the top of my grader cohort, producing excellent work with the highest volume.

I performed up to 300 emotion recognition tasks a day, which involved listening to 30-second audio clips and determining the speaker emotion, confidence, activation and identifying background noise. These clips often involved separate speakers and I had to distinguish between each speaker. I was ranked in the top of my grader cohort, producing excellent work with the highest volume.

2025 - 2025

Education

Work History