How I ended up here

I came into Duke planning to be a pre-med neuroscience major. Social media had affected my mental health a lot, and I wanted to understand why apps like Instagram and TikTok are so good at pulling people in, and I figured the answer was somewhere in the brain. (I also took one look at the Neuro 101 syllabus and changed my schedule, so it wasn't a purely noble pivot.)

Then I took a class called Googlization of Knowledge, about how tech companies design their products to keep you on them, and it changed everything. A lot of the answer I was looking for wasn't in neurons at all, it was in product decisions, and there was real work to be done on them outside of writing code.

I switched to public policy with a certificate in digital intelligence and took every tech ethics class I could find. My internships were a pretty random assortment, and most of them came last minute: I'd apply to a lot of things, hear nothing back, and then a coffee chat or a bio I posted in a Slack group would turn into a job.

They were all different, but they circled the same idea, which was less about regulating technology and more about building it responsibly. I worked with Facebook whistleblower Frances Haugen on a database of social media harms, helped policymakers in four Latin American countries turn cybersecurity guidance into legislation, and at Metaphysic, the generative AI company that de-aged Tom Hanks in Here, helped build the safety review process for clients whose faces were being altered by AI.

The job that changed my direction was at Tremau, a Paris-based trust and safety consultancy (I worked remotely from my childhood bedroom, so there were a lot of 4am calls). We assessed some of the largest social media platforms in the world under the EU Digital Services Act and the UK Online Safety Act: identify a risk, like child sexual abuse material or misinformation, then map the mitigation the platform already had or should add.

The work mattered and I loved it, but I kept feeling like I was checking boxes after the fact. A platform could pass every check and still have serious harm on it, because being compliant isn't the same as being safe. I wanted to get at the problem earlier, where the product actually gets designed.

The catch was that I'd taken one Python class freshman year and hated it. I also kept noticing that engineers and the policy and business people around them barely talk to each other, and mostly don't have the language to. So I figured I'd try to become all of it, and applied to Duke's AI engineering master's in March of my senior year, very late. It took a lot of conversations to convince them I actually liked math (I took linear algebra and calc 3 in high school, I just hadn't touched it since), and they let me in.

I had never used GitHub. I failed my first exam and thought about quitting basically every day for the first month. I didn't, and by the end I was shipping NLP pipelines, computer vision models and full-stack AI tools, usually as one of a handful of women in the room.

Now I'm an associate product manager at Visa in San Francisco, where I own go-to-market for a B2B payments product and build AI workflows for my team, and I'm the managing editor of The AI Collective, where I write The Stress Test, a weekly column about AI accountability.

The problems I care most about are still the ones I started with: how AI systems behave, who they affect, and how you'd know something was going wrong before it reached people. Most of what I build is some version of that, and I'd like to keep working on it for a long time (ideally with fewer 4am calls).

← Back to the short version