Rossana F.

Key Skills

Software

Don't disclose

Mindrift

Top Subject Matter

Search quality evaluation for generative/LLM-adjacent content ranking safety and relevance

AI safety evaluation and adversarial red teaming of LLM outputs

Creative writing and reasoning evaluation for generative AI training data quality

Top Data Types

Text

Document

Top Task Types

Red Teaming

Text Generation

Freelancer Overview

Analytical and detail-oriented professional with over 14 years of experience in data validation, content quality, and B2B communication within the automotive sector. My work focuses on high-complexity RLHF tasks and Red Teaming to identify model vulnerabilities and ensure safety in LLM outputs. As a Certified Google Search Quality Rater, I apply E-E-A-T standards and PQ/NM evaluations to improve search relevance and content credibility. My background includes over a decade of managing ISO 9001:2015 compliant processes in the B2B automotive sector, where I developed a strong mindset for audit and protocol design. With C1 English proficiency, I am dedicated to delivering accurate, rubric-compliant evaluations for global AI training projects.

Labeling Experience

Creative & Reasoning Tasks Specialist (Mindrift)

MindriftTextText Generation

Evaluated Italian- and English-language creative and long-form generation tasks, assessing coherence, cultural accuracy, stylistic quality, and instruction-following fidelity. Performed preference ranking and comparative evaluation of model outputs for creative writing, argumentation, and complex reasoning prompts. Contributed to training-data quality by producing high-quality reference responses and nuanced rationales across diverse creative domains. • Assessed creative writing and long-form generation quality • Ranked and compared responses for preference and instruction adherence • Authored reference responses and rationale supporting evaluation • Evaluated multilingual outputs (Italian and English) for QA quality

2023 - Present

RLHF & AI Safety Evaluator (Invisible Tech)

TextRed Teaming

Designed and executed systematic red teaming campaigns against frontier LLMs to identify jailbreak vectors, harmful output patterns, and policy boundary failures. Detected bias, hallucinations, and ethical violations in model outputs and provided actionable feedback to improve alignment and safety compliance. Produced evidence-based safety rationales aligned with project rubrics using a systematic, audit-minded approach. • Red teamed LLMs for jailbreak and policy boundary failures • Flagged harmful outputs, bias, and hallucinations with rationales • Performed intent and harm/policy boundary analysis per rubric • Ensured consistent, reproducible safety assessments through structured methods

2023 - Present

Search Engine Evaluator (RWS)

Don't discloseText

Performed Page Quality (PQ) and Needs Met (NM) evaluations for search results using Google Search Quality Rater Guidelines, producing rubric-aligned judgments. Applied the E-E-A-T framework to assess credibility, source authority, and relevance for Italian and English queries. Delivered structured, evidence-based rationales consistently across high-volume evaluation cycles. • Rated page quality and needs satisfaction for search tasks • Assessed E-E-A-T factors (credibility, authority, relevance) • Produced consistent written rationales per rubric requirements • Evaluated across Italian and English query contexts

2022 - Present

Education

Work History