Marisol Ramos

About Marisol Ramos

I’m an AI data and model QA analyst with 2+ years of experience evaluating large-scale…

Experience

1. Ai Behavior & Risk Analysis Technical Writer Meta Super Intelligence Labs

   Magnit Global

   Sep 2023 — Present · San Francisco, CA, US

   Led adversarial testing for LLM safety evaluations across multilingual contexts, simulating user personas (e.g, minors, regional dialects, non-native speakers) to identify model weaknesses and surface latent safety risks.• Developed and deployed red team strategies utilizing adversarial ML tactics including prompt injection, filter evasion, and obfuscation in Spanish and English, reducing high-harm model responses by ~37% across flagged categories.• Authored over 200+ high-impact safety reports informing cross-functional teams (engineering, policy, compliance, and marketing) with data-backed recommendations, influencing safety roadmap decisions.• Streamlined test documentation workflows and attack reporting standards, improving diagnostic report turnaround time by 25% and increasing successful exploit replication by internal QA teams by 60%.• Contributed to enhancements in RLHF/SFT alignment efforts by identifying nuanced prompt exploits and false refusals in model outputs, leading to the adoption of 3 new safety triggers

Education

• Carrington College California

  Health Information Technology

• Western Governors University

  Bachelor of Science - BS