Rachit D.

Key Skills

Software

Don't disclose

Top Subject Matter

Artificial Intelligence and Machine Learning education/evaluation

Coding and Computer Science

Mathematics

Top Data Types

Text

Document

Image

Top Task Types

RLHF

Evaluation/Rating

Classification

Transcription

Fine-tuning

Question Answering

Text Generation

Text Summarization

Freelancer Overview

My experience in AI training and data curation is built upon a foundation of rigorous academic research and diverse project-based application. During my M.Tech at IIT Hyderabad and throughout my work on various platforms, I have spearheaded complex initiatives ranging from the OpenClaw and Atlas projects to advanced mathematical and STEM-focused modeling. These projects required not only technical proficiency in model training and architectural optimization but also a meticulous approach to data quality, where I handled specialized datasets to ensure high precision in downstream tasks. This background allows me to understand the critical link between raw data integrity and the ultimate robustness of AI systems, whether I am fine-tuning transformer models or developing highly specific STEM algorithms. What sets me apart is my ability to translate highly technical, domain-specific requirements into structured, high-quality training data. My work on advanced STEM and mathematical projects has sharpened my ability to handle complex, logic-heavy datasets that require both deep subject-matter expertise and strict attention to structural consistency. Having solved over 400 algorithmic challenges, I bring a disciplined, analytical approach to every labeling task, emphasizing edge-case detection and logical clarity. By combining the precision of a developer with the analytical depth of a researcher, I am well-equipped to contribute to diverse, high-cognitive-load projects that demand both accuracy and a sophisticated understanding of AI training workflows.

Labeling Experience

Whitebeard

TextRLHF

In this project, I served as an expert domain evaluator for the Whitebeard workflow, where my primary responsibility was to assess and refine AI model responses to complex technical prompts. I thoroughly analyzed paired model outputs across several advanced mathematical disciplines, including multivariable calculus, complex analysis (specifically Hadamard finite-part regularization), and combinatorics. For each task, I evaluated the prompts for ratability based on strict guardrails regarding safety, PII, and data completeness. I then performed a comparative analysis to select the preferred response, judging dimensions such as instruction following, technical accuracy, and structural layout. Finally, I remediated formatting failures—such as broken LaTeX delimiters or syntax errors in code blocks—to generate a polished, 'gold standard' final rewrite suitable for platform deployment.

2026 - 2026

OpenClaw Atlas

TextRLHF

My recent project involved evaluating an LLM agent’s ability to securely process incoming data from an archive while maintaining precise internal tracking logs. The agent was designed to filter premium client updates within a strict 12-hour window while enforcing state records to completely block duplicate processing. To prioritize user privacy, I implemented rigorous scrubbing protocols that masked personal identifiers like phone numbers with specialized tags, while ensuring sensitive email data was completely omitted from the output. Finally, I benchmarked the model's security by testing its behavioral resistance against malicious prompt injections hidden within the dataset, specifically designed to attempt the leakage of raw, unredacted files.

2026 - 2026

Education

Work History