Rahel Paulos

Key Skills

Software

Mindrift

CVAT

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Image

Text

Video

Top Label Types

Text Summarization

RLHF

Red Teaming

Prompt Response Writing SFT

Segmentation

Classification

Freelancer Overview

I am a Software Engineering graduate with a focus on AI system evaluation and "Human-in-the-Loop" training. My experience includes multimodal data labeling across platforms like Mindrift (Toloka AI), and Clickworker, where I’ve handled everything from image/video annotation for autonomous systems to evaluating LLM reasoning and prompt engineering

Labeling Experience

LLM Reasoning and Factuality Evaluation

MindriftTextText SummarizationRLHF

Performed Reinforcement Learning from Human Feedback (RLHF) to improve Large Language Model (LLM) accuracy. Tasks included evaluating model-generated responses for logical consistency, factual precision, and safety compliance. I identified edge cases in reasoning and refined prompts to test model boundaries. My focus was on providing nuanced feedback to reduce hallucinations and ensure the model followed complex multi-step instructions.

Performed Reinforcement Learning from Human Feedback (RLHF) to improve Large Language Model (LLM) accuracy. Tasks included evaluating model-generated responses for logical consistency, factual precision, and safety compliance. I identified edge cases in reasoning and refined prompts to test model boundaries. My focus was on providing nuanced feedback to reduce hallucinations and ensure the model followed complex multi-step instructions.

2025 - 2025

Multimodal Data Curation for Autonomous Navigation

CVATImageSegmentationClassification

Curated and validated over 1,000 multimodal training sequences for autonomous robot navigation systems. I applied nuanced judgment to classify complex terrain features and obstacles that were frequently missed by automated pre-classification. My work directly improved dataset consistency and reliability for human-in-the-loop (HITL) pipelines, specifically focusing on identifying subtle environmental hazards.

Curated and validated over 1,000 multimodal training sequences for autonomous robot navigation systems. I applied nuanced judgment to classify complex terrain features and obstacles that were frequently missed by automated pre-classification. My work directly improved dataset consistency and reliability for human-in-the-loop (HITL) pipelines, specifically focusing on identifying subtle environmental hazards.

2024 - 2025

Education

Work History