Zahra Khan

Key Skills

Software

Other

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Label Types

Evaluation Rating

RLHF

Text Generation

Text Summarization

Freelancer Overview

I specialize in building high-quality training data for reasoning models, with a focus on biomedical and clinical content. At Outlier, I design adversarial prompts that intentionally expose failure modes (hallucinations, shallow chain-of-thought, instruction drift), then create gold references and MECE, atomic rubrics to evaluate pointwise and pairwise outputs. My work spans reference-grounded tasks (e.g., RCTs, physiology, neurology, ethics/REB language), safety-sensitive QA (avoiding PHI, enforcing citation discipline), structured outputs (JSON/spec validation), and multimodal/text-to-text instructions. I routinely do error taxonomy design, response rating, and editorial rewrites to convert strong drafts into publication-quality answers. What sets me apart is the blend of domain depth (ICU trial coordination, biomarker methods, GCP/REB literacy) and data quality rigor: I translate dense scientific sources into precise labeling specs, build rubrics that reveal model weaknesses (not just score them), and deliver datasets that measurably improve reasoning, faithfulness, and instruction following.

Labeling Experience

Gen LLM Trainer

Scale AITextText GenerationRLHF

Various projects

Various projects

2024

Education

Work History