Visiting Researcher
Sep 2026 - Present
Building a benchmark that separates genuinely contested litigation questions from settled ones, measuring whether models collapse doctrinal disagreement into a single answer. Expert annotation with practicing litigators, with Prof. Julian Nyarko and Dr. Megan Ma.
Building a benchmark that tests whether AI legal tools flatten genuine doctrinal disagreement into one confident answer, a risk for how litigators and courts come to rely on them. Expert annotation with practicing litigators, with Prof. Julian Nyarko and Dr. Megan Ma.
Lawyers sometimes truly disagree, but AI tends to pick one answer and sound sure about it. I’m building a test for whether AI can tell “lawyers disagree” from “everyone agrees,” with real litigators helping.
Benchmark separating contested from settled litigation questions; tests whether LLMs collapse doctrinal disagreement. Litigator-annotated.
Research Assistant
University of CambridgeDept. of Computer Science and Technology · Centre for Human-Inspired AI
July 2026 - Present
Construct validity in AI bias evaluation (continued dissertation work). Dark patterns in interactive systems with Dr. Katie Seaborn (Aspire Lab). Ethics-centered hackathons as a method for participatory design with Dr. Malak Sadek (HIVE Lab).
Construct validity in AI bias evaluation: do the measures developers and regulators rely on actually measure the bias they claim to? Continued from my dissertation, alongside work on dark patterns and ethics-centered hackathons for participatory design.
When we test an AI for bias, the test itself can be wrong. I study whether bias tests really measure bias. I also look at sneaky app tricks, and how hackathons help people design fairer tech.
Construct validity of AI bias evaluation; dark patterns; ethics-centered hackathons as participatory design.
Researcher
PluralisticAI Lab Dr. Zhenzhen Qi (University of Connecticut) · Dr. Avijit Ghosh (Hugging Face)
Mar 2026 - Present
Co-designing a controlled human-subjects study evaluating whether an open-source, multi-persona LLM interface can reduce political polarization. Combining cognitive science, political studies, AI alignment, game design and HCI.
Co-designing an open-source, multi-persona LLM interface and the controlled human-subjects study testing whether it reduces political polarization. Spans game design, HCI, cognitive science and AI alignment.
Online, people end up angry at each other. I’m helping test a chatbot that speaks in many different voices, to see if it helps people disagree less.
Human-subjects study: open-source multi-persona LLM interface vs. political polarization.
See more →
Civic, Open & Resilient Democracy Action Fellow
Odyssean Institute Jacob Haimes (Apart Research) · Dr. Swaptik Chowdhury (RAND)
Apr 2026 - Sep 2026
Built a 297-variable causal loop diagram from a 23-expert AI risk horizon scan, plus an LLM coding pipeline matching inter-human agreement on risk-domain classification. Outputs feed decision support for UK and global citizens' assemblies.
Translated expert views on AI risk into causal loop diagrams for an AI governance horizon scan, informing the decision-support tools used in the 2026 UK assembly.
I read what experts think could go wrong with AI, then drew maps of how those things cause each other, to help leaders in the UK assembly decide.
Coded expert responses into causal loop diagrams for an AI governance horizon scan; feeds 2026 UK assembly tools.
Robotics Software Engineering Intern
May 2025 - Aug 2025
Designed an interactive dashboard and workflow for managing autonomous robot power systems, combining real-time battery feedback, predictive alerts, and manual override controls. Iteratively prototyped and user-tested for trust, reliability, and operator agency.
Built an interactive dashboard and workflow for managing autonomous robot power systems: real-time battery feedback, predictive alerts, and manual override controls. Iteratively prototyped and user-tested.
Designed a dashboard and workflow for robot power management, balancing automation with operator control. Prototyped and user-tested iteratively for trust, reliability, and operator agency.
A robot runs on a battery. I made a screen that shows how much is left, warns you before it runs out, and lets a person take over. Then I asked real people if they trusted it.
Robot power dashboard: live battery, predictive alerts, manual override. User-tested.
Research Assistant
University of WashingtonWordplay Project · UW Information School · Dr. Amy Ko
Sep 2024 - June 2025
Led template design for an accessible, multilingual programming language learning platform deployed in local schools, scaffolding first-time coders with pre-built “templates”. Coordinated an open-source contributor community.
Led deployment of an accessible, multilingual programming language learning platform in local schools. Coordinated an open-source contributor community and built a pre-built “templates” system for first-time coders.
Led template design for a multilingual programming learning platform deployed in local schools, scaffolding first-time coders.
Led template design for an accessible, multilingual learning platform deployed in local schools, shaping a “templates” system around how first-time coders get stuck.
I helped build a way for kids to learn coding in many languages, and brought it to local schools. I also helped volunteers work together and made starter templates so beginners don’t get stuck.
Led rollout of accessible multilingual coding platform in local schools; ran open-source community; built templates.
See photo →
Backend Software Engineering Intern
HowellCoStealth Startup
Jul 2024 - Sep 2024
Contributed to backend service development at a beta-stage startup, designing products, devising a testing strategy, and using Azure services and Python scripting to enhance backend infrastructure on the mobile app.
Designed products and devised a testing strategy at a beta-stage startup, while using Azure services and Python scripting to strengthen the mobile app’s backend.
I helped a brand-new startup’s phone app work better behind the scenes. I planned how to test it and used Azure and Python to make it stronger.
Backend and product work at a beta-stage startup; testing strategy; Azure and Python.
Software Engineering Intern
Code.orgProduct-Oriented EdTech Engineering
Jul 2023 - Sep 2023
Engineered full-stack, web-based coding games in React and Ruby on Rails for a platform used by 80 million students and 2 million teachers, working across engineering and product.
Engineered web-based coding games for a platform used by 80 million students and 2 million teachers, working across engineering and product.
I made coding games that kids play to learn programming. Millions of students and teachers use them.
Full-stack coding games in React and Rails for a platform used by 80M students, 2M teachers.
Operations & Product Intern
FiveableEdTech Startup
Apr 2021 - Nov 2021
Scaled internal operations and product workflows during the company’s transition into a consumer platform. Led design of a student-facing platform aimed at improving access for first-generation and low-income learners.
Led design of a student-facing platform aimed at improving access for first-generation and low-income learners, during the company’s move into a consumer product.
A small education company was growing up. I helped keep its daily work running, and helped build a site for students who are first in their family to go to college or don’t have much money.
Scaled ops and product workflows; led design of student platform for first-gen and low-income learners.
