Research & Strategy for User-Centered AI
My foundation in linguistics and cross-cultural communication gives me a unique lens for solving today's challenges in AI. Over 7 years across Google Assistant and Google's AI product Gemini (Bard), my work has grown to include building adversarial test sets from scratch to proactively identify safety risks, conducting market-specific research to ensure cultural relevance on a global scale, and designing new quality rubrics based on core UX principles.
Areas of Expertise
Global Experience
My professional experience is complemented by rich international engagement, including teaching and language immersion in Japan on the JET Program, and graduate studies in linguistics at the University of Oxford. Independent studies of language, art, and history across Europe have given me a firsthand appreciation for diverse cultural frameworks, which is central to my work building truly globally-aware AI products.
Select a point on the map to see details.
Portfolio Highlights
Here are a few projects that I'm particularly proud of, as they represent key moments in my journey of applying linguistic theory to real-world challenges.
Gemini Global Model Evaluation
Role: Engineering Linguist, Google
Since March 2023, I've been leading evaluation strategy for multilingual quality and feature launches for Gemini (Google Bard). My team is the final checkpoint for model evaluation and tuning before any global feature launch, including Gemini Live, Actions, and model releases for Gemini 2.5 Flash and 2.5 Pro.
Methodologies:
Cultural Context Adaptation
Role: Senior Linguist, Google
I developed and implemented a pilot program to evaluate and improve Gemini's cultural relevance and sensitivity, focusing on markets like India, Brazil, and Japan. We discovered, for instance, that early models often lacked the expected levels of formality and honorifics, which are crucial for polite conversation in many cultures. This involved designing specific evaluation tasks and collaborating with UX researchers on in-market studies to inform model training.
Methodologies:
Conversation Design for Conciseness
Role: Conversation Designer, Google
When Gemini Live was introduced, a key challenge was ensuring responses felt natural in a spoken conversation—nobody wants to hear a long, verbose answer. I ran multiple experiments evaluating the model with different conciseness parameters, validated responses per language, and processed rater feedback to fine-tune the verbal output, while also coordinating it with the mirrored text on the app's interface.
Methodologies:
The MetaNet Project
Role: Semantic Analyst, ICSI, Berkeley
As part of an IARPA-funded initiative, I helped build a large-scale, multi-lingual database for detecting metaphorical language automatically. My role was to create the semantic taxonomies and grammatical resources needed for the system to identify non-literal language in massive text corpora. This was a formative experience in bridging theory and computation, and I'm proud that the work continues today in the MetaNet project at UBC, now led by my long-time collaborator Dr. Elise Stickles.
Methodologies:
"Metaphor in Grammar" Research
Role: Ph.D. Candidate, UC Berkeley
My doctoral dissertation was the culmination of my academic training, where I argued that conceptual metaphor is a core principle of grammar. I analyzed thousands of real-world sentences from the FrameNet database and other custom corpora to show how abstract concepts are structured and understood through more concrete, embodied metaphors.
Methodologies:
Medical & Pharmaceutical Localization
Role: Project Manager, Corporate Translations Inc.
Before my work in tech, I was a Project Manager specializing in pharmaceutical and medical translations. In this role, I was the last person to spot and correct critical mistakes and the first to field client feedback. I managed quality control for localization of sensitive materials like quality-of-life questionnaires, pharma trial documents, and ingredient inserts. This involved hands-on editing, proofreading, and quality assurance across all language pairs, and gave me a deep appreciation for the precision required in high-stakes localization projects.
Methodologies:
Publications
| Year | Title | Summary | Type | Citation |
|---|
Awards & Fellowships
- 2023: Google Assistant Tech Impact Award
- 2017: University of California, Merced, Center for the Humanities Seed Grant (with Teenie Matlock and Dalia Magaña): "Cancer metaphors in Spanish health stories" ($1,900, Co-PI)
- 2016-2018: University of California Chancellor's Postdoctoral Fellowship
- 2016: Linguistic Society of America (LSA) Student Abstract Award
- 2015: Digital Humanities Summer Institute Certificate, UC Berkeley
- 2013-2014: Doctoral Completion Fellowship, UC Berkeley
- 2013: Outstanding Graduate Student Instructor Award, UC Berkeley
- 2012: The Gen Foundation Grant
- 2010-2011: A. Richard Diebold, Jr. Fellowship, UC Berkeley