Blank Z.

Key Skills

Software

No software listed

Top Subject Matter

STEM (Automotive Engineering, Fluid Mechanics/CFD, Thermodynamics) and Financial Markets

Engineering research

technical writing

Top Data Types

Text

Top Task Types

RLHF

Freelancer Overview

Final-Year Project Researcher (Wind Energy). Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Outlier AI, Claude, and GPT-4. Education includes Bachelor of Engineering, Elizade University (2026). AI-training focus includes data types such as Text and labeling workflows including RLHF, Evaluation, and Rating.

Labeling Experience

Data Annotator / AI Trainer — Outlier AI (Remote, 2024 – Present)

TextRLHF

Performed AI response evaluation on Outlier AI, rating model outputs for accuracy, helpfulness, coherence, and instruction-following to support RLHF pipelines. Wrote and evaluated prompts for technical STEM topics by verifying factual correctness, reasoning quality, and hallucination presence using engineering domain knowledge. Created preference labels by comparing paired AI responses and providing written rationales selecting the better response against task criteria. • Rated responses with rubric-based consistency across safety, accuracy, and adherence dimensions • Provided justification text to explain why one response better met the stated criteria • Adapted to new project types and updated guidelines while maintaining reproducible decisions • Maintained high precision with a zero-tolerance approach to factual errors

2024 - Present

Independent LLM Evaluator / Prompt Evaluation — Daily Practice (2023 – Present)

Text

Used frontier and local LLMs daily to identify and document hallucinations, reasoning errors, and factual inaccuracies in STEM and financial-domain outputs. Tested and compared locally-run open-source model outputs against frontier models to understand how model size and training affect quality. Designed structured prompts and few-shot examples to elicit more accurate, instruction-following responses. • Calibrated evaluation based on strengths and failure modes of Claude, GPT-4, and Gemini • Documented quality issues suitable for reuse as labeling criteria for future tasks • Compared outputs from Ollama and GPT4All against frontier models for error analysis • Iteratively refined prompt formats and examples to improve response reliability

2023 - Present

Education

Work History