I'm an associate product manager at Visa and the managing editor of The AI Collective. I have a master's in AI from Duke, after starting out in public policy and trust and safety.
Selected work
Featured case · Duke, spring 2026
With Shreya Mendi and Andrew Jin, I built an NLP pipeline over 1.4 million DOJ documents from the Epstein files to measure the gap between how much documentary evidence exists about each person named and whether anything ever happened to them.
Evidence against consequence, one tick per person. 47 people score in the top half of the evidence scale and have no recorded consequence.
No recorded consequence546
Soft consequence: resigned, sued, reputational damage22
Hard consequence: arrested, convicted3
Equity audit
An audit of AI-driven transportation planning in Durham, NC, built around four bias tests. It found roughly 25% undercounting in low-income areas and a 29.5% disparate impact ratio in how infrastructure gets allocated.
Content analysis
Scores psychological manipulation in TikTok videos. It transcribes each video with Whisper, then scores it with a fine-tuned DistilBERT classifier, using Claude as an LLM judge for labeling.
Computer vision
Finds people in thermal drone footage for search and rescue, using Faster R-CNN with temperature-scaling calibration so the model's confidence means something.
Compliance tooling
Browser extension
A privacy-first Chrome extension that tracks the carbon footprint of your web browsing in real time.
Experience
Sept 2026 – now
Associate Product Manager · Visa, rotational APM program, San Francisco
I own go-to-market for a B2B payments product at Visa: the pricing model, partner integrations, and getting engineering, legal and compliance aligned from API spec to launch, while working with VP and SVP leadership and external partners. I also build LLM agent workflows for the Core Products team.
May 2026 – now
Managing Editor, Editorial Standards · The AI Collective
I run editorial standards for a 187K‑subscriber AI newsletter and write The Stress Test, a weekly AI accountability column.
June – Aug 2026
Research Manager · Deep Tech at Duke
Built a full-semester tech policy course in 9 weeks, including 8 case studies of policies that failed (SB 1047, the Colorado AI Act, NYC Local Law 144) and a 16‑mode failure taxonomy.
Feb 2025 – May 2026
Research Fellow · Deep Tech at Duke
Designed the measurement method for a study of AI-generated music on Spotify, and led an AI agent risk assessment that Paladin adopted into its published risk framework.
May – Sept 2024
Trust & Safety Analyst · Tremau
EU Digital Services Act risk assessments for a major social media platform, mapping 15+ risk pathways into mitigation roadmaps and escalation workflows.
Feb – July 2024
AI Ethics, Policy & Research Fellow · Metaphysic.ai
Co-developed a safety review framework for hyper-realistic generative AI at a point when no industry benchmarks existed, and onboarded 8 clients with it in 3 months.
Dec 2023 – May 2024
Cybersecurity Research Assistant · Duke Sanford School of Public Policy
Helped government and private-sector leaders in Panama, Chile, Colombia and Costa Rica turn cybersecurity best practices into legislation.
May – Aug 2023
Tech & Public Policy Fellow · Beyond the Screen
Worked with Facebook whistleblower Frances Haugen to build a database of social media harms, and wrote 5 literature reviews on moderation, misinformation and polarization.
2024 – 2025
Co-founder & VP · Duke Tech for Change, and Board Member · All Tech Is Human
Writing
Why AI companions feel supportive, reassuring and human, and why that's exactly the point.
Governance challenges of invasive brain-computer interfaces.
What Unmasking AI revealed about datasets, omissions and the limits of post-hoc explanations.
What child development and psychopathy can teach us about building safer AI systems.
Why deepfake abuse is spreading faster than our protections.
Applicant tracking systems have reshaped hiring, but at what cost to fairness and equity?
What Flo Health teaches us about privacy in an age of digital surveillance.
The hidden risks of facial recognition in everyday life.
How Apple, Google, Amazon, Microsoft and Meta rewrote competition, and why regulators are still playing catch-up.
How thinking like a machine learning model helped me make a major career change.
About
I started on the policy side, studying how technology shapes society and where the guardrails fail, which took me to trust and safety work at Tremau and AI ethics at Metaphysic. As AI started moving faster than regulation, I realized the real decisions get made inside the product cycle, so I learned to build.
Most of what I make follows the same pattern: find a system that makes decisions about people, measure what it actually does, and put the gap somewhere people can see it. I hold an MEng in AI and a BA in Public Policy from Duke.
The longer version, from neuroscience to now →
Evals and LLM-as-judge labeling · NLP pipelines · agent workflows · AI risk assessment · Python, TypeScript, PyTorch, Hugging Face, FastAPI, GCP