Building safer, more trustworthy products at the intersection of AI, data, and human behavior.
TRUST & SAFETY | AI PRODUCT | USER RESEARCH | RESPONSIBLE AI
|
|
An Advocate for Responsible Innovation
How do we build technology people can trust without losing sight of why they want to use it in the first place?
My work spans Trust & Safety, LLM evaluation, user research, analytics, and product strategy. I’ve designed evaluation systems for AI products used by educators and students, investigated failures in automated safety measurement, studied bias and user behavior in recommendation systems, built analytics products used by executives, and worked across Product, Engineering, Policy, Legal, and Data Science to turn complex findings into product decisions. I’m especially interested in the problems that emerge when technology meets real human behavior: how we measure safety, how we distinguish genuine risk from false positives, how protections change for younger users, how language evolves faster than our systems, and how we build guardrails without designing the value out of a product.
How I Approach Trust & Safety
Understand the user, not just the violation.
Safety systems operate within human contexts.
I use behavioral research, user research, and real-world usage patterns to understand the people behind the metrics.
Measure what actually matters.
A safety score is only useful if the measurement behind it is trustworthy.
My work includes KPI development, LLM evaluation, false-positive and false-negative analysis, experimental design, and validation.
Turn findings into product decisions.
Research has limited value if it ends with a report. I work across technical and non-technical teams to translate evidence into requirements, priorities, interventions, and measurable outcomes.
Build safety into the product.
I’m interested in Trust & Safety as a product discipline—not simply enforcement after something goes wrong. The best safety systems consider user experience, product growth, technical constraints, and risk together.
My work spans Trust & Safety, LLM evaluation, user research, analytics, and product strategy. I’ve designed evaluation systems for AI products used by educators and students, investigated failures in automated safety measurement, studied bias and user behavior in recommendation systems, built analytics products used by executives, and worked across Product, Engineering, Policy, Legal, and Data Science to turn complex findings into product decisions. I’m especially interested in the problems that emerge when technology meets real human behavior: how we measure safety, how we distinguish genuine risk from false positives, how protections change for younger users, how language evolves faster than our systems, and how we build guardrails without designing the value out of a product.
How I Approach Trust & Safety
Understand the user, not just the violation.
Safety systems operate within human contexts.
I use behavioral research, user research, and real-world usage patterns to understand the people behind the metrics.
Measure what actually matters.
A safety score is only useful if the measurement behind it is trustworthy.
My work includes KPI development, LLM evaluation, false-positive and false-negative analysis, experimental design, and validation.
Turn findings into product decisions.
Research has limited value if it ends with a report. I work across technical and non-technical teams to translate evidence into requirements, priorities, interventions, and measurable outcomes.
Build safety into the product.
I’m interested in Trust & Safety as a product discipline—not simply enforcement after something goes wrong. The best safety systems consider user experience, product growth, technical constraints, and risk together.
|
Selected Work
AI Safety & Evaluation — MagicSchool AI Measured safety and quality across 130+ AI tools and features, designed 25+ evaluation studies, built 11 safety and quality KPI dashboards, and identified model and system failures that contributed to ~270 corrective actions. When Evaluators Cry Wolf — LLM Evaluation Research Investigated false-positive behavior in LLM-as-judge systems and developed approaches for improving evaluator signal quality through prompt enhancement and ensemble selection. Recommendation Systems & Algorithmic Bias — TikTok Research Led a 100-participant study generating 3,000+ data points to examine user experience and potential disparities within TikTok’s recommendation ecosystem. Youth Safety & Digital Wellbeing — CU Boulder Research Studied how developing users interact with technology and contributed research and educational resources focused on healthier, more intentional relationships with digital products. OfferHopper — AI Product Development Designed and built an AI-assisted career platform that turns years of professional-development curriculum and mentorship into an interactive resource students can continue using independently. |
Core Expertise
Trust & Safety Product Product Strategy · Safety by Design · Risk Identification · Product Quality · Cross-Functional Leadership · KPI Development · Roadmapping AI Safety & Evaluation LLM Evaluation · Responsible AI · Prompt & Context Engineering · Synthetic Testing · Evaluator Design · False Positive/Negative Analysis · Quality & Accuracy Reviews Users & Research User Research · Human-Centered Design · Mixed-Methods Research · Behavioral Analysis · Usability Testing · Survey Design · Algorithmic Bias Data & Experimentation SQL · Python · Experiment Design · A/B Testing · Statistical Sampling · Tableau · Power BI · Performance Monitoring |