Arnav Singh

Key Skills

Software

Labelbox

Scale AI

Mercor

Mindrift

Top Subject Matter

LLM Fine-Tuning (Coding, Math, Reasoning, Creative Writing)

Multimodal Annotation (Image + Text Understanding)

Audio & Speech Data Annotation

Top Data Types

Image

Video

Audio

Top Task Types

RLHF

Prompt Response Writing SFT

Computer Programming Coding

Freelancer Overview

I’ve worked on AI training and data labeling through platforms like Outlier AI and Mercor, where I spent time reviewing and improving AI-generated responses. My role mainly involved checking for accuracy, clarity, and logical consistency, as well as ranking different outputs and giving feedback to make the models better. Over time, I got familiar with how AI can sometimes make mistakes like giving incorrect information or vague answers and learned how to spot and fix those issues effectively. Along with this, I have a background in Computer Science, which helps me approach tasks in a structured and analytical way. I pay close attention to detail and try to keep my work consistent with the given guidelines. I’m comfortable adapting to different types of tasks and enjoy working on improving AI systems, which makes me a dependable contributor in data labeling and training projects.

Labeling Experience

AI Trainer & Data Contributor

LabelboxVideoRLHF

As an AI Trainer and Data Contributor for Mercor, and Scale AI, I completed a high volume of reinforcement learning from human feedback (RLHF) and evaluation tasks for frontier language model fine-tuning. My responsibilities included RLHF preference labeling, instruction-following evaluation, code correctness review, and creative-writing quality scoring. Additionally, I authored prompts across coding, math, and reasoning domains, performed factual verification, and conducted entity annotation using Labelbox software. • Completed over 500 AI training tasks emphasizing RLHF preference labeling and output evaluation. • Maintained a >95% quality acceptance rate and met strict deadlines for multiple concurrent annotation projects. • Utilized Labelbox, Outlier, Mercor, and Scale AI software for evaluation and annotation across diverse domains. • Specialized in preference labeling, factual verification, and creative writing quality scoring.

As an AI Trainer and Data Contributor for Mercor, and Scale AI, I completed a high volume of reinforcement learning from human feedback (RLHF) and evaluation tasks for frontier language model fine-tuning. My responsibilities included RLHF preference labeling, instruction-following evaluation, code correctness review, and creative-writing quality scoring. Additionally, I authored prompts across coding, math, and reasoning domains, performed factual verification, and conducted entity annotation using Labelbox software. • Completed over 500 AI training tasks emphasizing RLHF preference labeling and output evaluation. • Maintained a >95% quality acceptance rate and met strict deadlines for multiple concurrent annotation projects. • Utilized Labelbox, Outlier, Mercor, and Scale AI software for evaluation and annotation across diverse domains. • Specialized in preference labeling, factual verification, and creative writing quality scoring.

2025 - Present

AI Annotator

ImagePrompt Response Writing SFT

Worked on image-based AI training tasks where I created and evaluated prompt–response pairs based on visual inputs. The role involved carefully analyzing images and writing clear, accurate, and context-aware responses that align with the given prompts. I ensured that the outputs captured important visual details, avoided assumptions, and remained factually grounded in what was visible in the image. In addition to writing responses, I reviewed and improved existing outputs by checking for clarity, relevance, and correctness. The work required strong attention to detail, the ability to interpret visual information accurately, and adherence to strict annotation guidelines. Quality was maintained through consistent self-review, following detailed rubrics, and ensuring that each response met the platform’s standards for accuracy and coherence.

Worked on image-based AI training tasks where I created and evaluated prompt–response pairs based on visual inputs. The role involved carefully analyzing images and writing clear, accurate, and context-aware responses that align with the given prompts. I ensured that the outputs captured important visual details, avoided assumptions, and remained factually grounded in what was visible in the image. In addition to writing responses, I reviewed and improved existing outputs by checking for clarity, relevance, and correctness. The work required strong attention to detail, the ability to interpret visual information accurately, and adherence to strict annotation guidelines. Quality was maintained through consistent self-review, following detailed rubrics, and ensuring that each response met the platform’s standards for accuracy and coherence.

2024 - Present

AI annotator and trainer

AudioRLHF

Worked on RLHF-based audio data annotation projects, focusing on improving the quality of AI-generated speech and language outputs. The role involved listening to audio samples, evaluating their accuracy, clarity, and naturalness, and ranking multiple responses based on defined quality criteria. I also provided detailed feedback to help refine model behavior and improve alignment with human expectations.

Worked on RLHF-based audio data annotation projects, focusing on improving the quality of AI-generated speech and language outputs. The role involved listening to audio samples, evaluating their accuracy, clarity, and naturalness, and ranking multiple responses based on defined quality criteria. I also provided detailed feedback to help refine model behavior and improve alignment with human expectations.

2025 - 2025

Education

Work History