Harsh D.

Key Skills

Software

Snorkel AI

Don't disclose

Top Subject Matter

Large language model training data for generalist and medical/healthcare domains

LLM-generated code evaluation for technical correctness

Medical/healthcare entity extraction and clinical text understanding for RLHF

Top Data Types

Text

Document

Top Task Types

RLHF

Red Teaming

Freelancer Overview

Remote AI Training Specialist & Expert Evaluator (Snorkel AI & Mercor). Brings 12+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Snorkel AI and Don't disclose. Education includes Master of Management of Information Systems, University of Nebraska Omaha (2016) and Bachelor of Engineering in Information Technology, Mumbai University (2012). AI-training focus includes data types such as Text, Computer Code, and Programming and labeling workflows including Evaluation, Rating, and Computer Programming.

Labeling Experience

Red Teaming & Safety Assessment (Terminus-2nd-Edition-Assessment)

Don't discloseTextRed Teaming

Assessed edge-case responses for safety, technical accuracy, and structural coherence in specialized technology and medical queries. Tested model boundaries using assessment workflows to identify failure modes and improve robustness. Provided evaluation outcomes to support safer and more reliable training behavior. • Edge-case response testing and safety assessment • Technical accuracy and structural coherence checks • Boundary testing in specialized technological/medical queries • Failure-mode identification for training improvement

2026 - Present

Data Adjudication & Refinement (EK-Mojave-Adjudicator-Quiz, EK-Mojave-Refinement-V2)

Snorkel AIText

Adjudicated and refined high-stakes AI training sets by resolving conflicts between primary annotators and reviewers. Enforced rubric adherence and ground-truth accuracy for quiz and refinement workflows. Conducted quality control transitions into adjudicator roles to improve dataset consistency. • Conflict resolution between annotators and reviewers • Rubric adherence and ground-truth enforcement • Adjudication for quiz and refinement datasets • High-stakes quality control and dataset consistency

2026 - Present

Generalist & Domain-Specific RLHF (EK-Mojave Series)

Snorkel AIRLHF

Performed reinforcement learning from human feedback (RLHF) on complex, multi-turn prompts in domain-specific settings. Extracted, classified, and verified medical terminology and clinical information to support healthcare query handling. Enforced zero-hallucination compliance through careful validation of extracted and generated content. • Multi-turn RLHF execution for healthcare prompts • Medical terminology and clinical text extraction/verification • Classification and domain-specific information validation • Zero-hallucination compliance checks

2026 - Present

Coder-Expert Assessment (Snorkel AI & Mercor)

Don't disclose

Assessed LLM-generated code outputs across multiple programming languages against strict software engineering standards. Verified program logic, algorithmic efficiency, and syntax to align model responses with production-grade requirements. Used structured evaluation criteria to determine correctness and completeness of generated code. • Code logic verification and correctness checks • Syntax and algorithmic efficiency validation • Cross-language comparison and evaluation • Quality gating for production alignment

2026 - Present

Remote AI Training Specialist & Expert Evaluator (Snorkel AI & Mercor)

Snorkel AIText

Evaluated, refined, and adjudicated LLM training data through submission, review, and adjudication tiers to maintain high accuracy and low defect rates. Applied rubric-based assessment and fact-checking to ensure outputs matched ground truth for advanced AI training sets. Resolved annotation conflicts to enforce consistent adjudication and quality control. • Rubric-based evaluation and fact-checking • Multi-turn dialogue assessment and refinement • Tiered submission/review/adjudication workflow • Cross-team adjudication and conflict resolution

2026 - Present

Education

Work History