Usman Salman

Key Skills

Software

Label Studio

Labelbox

Mindrift

Google Cloud Vertex AI

Micro1

Other

Internal/Proprietary Tooling

Top Subject Matter

LLMs – Prompt Engineering, Code Reasoning & Model Evaluation

Multimodal AI – Web Search Reasoning & Dataset Generation

STEM & Educational AI – Science Question Annotation

Top Data Types

Text

Image

Audio

Top Task Types

Object Detection

Text Summarization

RLHF

Fine Tuning

Evaluation Rating

Computer Programming Coding

Data Collection

Prompt Response Writing SFT

Function Calling

Text Generation

Classification

Segmentation

Entity Ner Classification

Freelancer Overview

I am an AI Trainer and Data Annotation Specialist with 3+ years of experience working on large-scale AI training and dataset development projects aimed at improving the reasoning, coding, and multimodal capabilities of modern language models. My work focuses on designing adversarial prompts, evaluating model responses, and writing high-quality reference answers to support reinforcement learning from human feedback (RLHF) pipelines. I have experience working on tasks related to code generation, debugging, API usage, testing, and code refactoring while assessing model outputs for correctness, reasoning quality, and instruction adherence. I have also contributed to multimodal dataset creation and annotation projects involving web search reasoning, educational STEM datasets, and structured problem-solving workflows. My experience includes working with annotation platforms such as Label Studio and Labelbox, along with structured data management using Google Sheets, JSON datasets, and Python-based preprocessing workflows. I specialize in creating high-quality, well-structured training data that helps AI systems better understand human reasoning, real-world problem solving, and multimodal information retrieval.

Labeling Experience

AI – Machine Learning Eng. (AI Trainer)

Label StudioTextRLHF

As an AI Trainer at Turing, I performed model response evaluation, prompt design, and ideal answer creation for training and aligning large language models. I labeled datasets for reasoning depth, correctness, instruction adherence, and robust stepwise solutions in multimodal LLM tasks. My work included annotation for STEM education, GAIA multimodal search, and structured code prompt datasets over diverse domains. • Designed adversarial prompts and evaluated LLM outputs for code generation and debugging tasks. • Generated and organized 5,000+ dataset entries with reasoning chains and human-authored solutions. • Annotated stepwise solution paths for Google STEM and created multimodal search queries with supporting evidence. • Managed annotation workflows in Labelbox and Label Studio, and conducted error analysis for data quality.

As an AI Trainer at Turing, I performed model response evaluation, prompt design, and ideal answer creation for training and aligning large language models. I labeled datasets for reasoning depth, correctness, instruction adherence, and robust stepwise solutions in multimodal LLM tasks. My work included annotation for STEM education, GAIA multimodal search, and structured code prompt datasets over diverse domains. • Designed adversarial prompts and evaluated LLM outputs for code generation and debugging tasks. • Generated and organized 5,000+ dataset entries with reasoning chains and human-authored solutions. • Annotated stepwise solution paths for Google STEM and created multimodal search queries with supporting evidence. • Managed annotation workflows in Labelbox and Label Studio, and conducted error analysis for data quality.

2024 - Present

Education

Work History