AI Product · Multi-Agent Systems · Agentic Web Infrastructure
Translating between people, technology, and product.
Design research and practice at Harvard GSD, building infrastructure for the agentic web at MIT. I translate between what people struggle with, what technology can enable, and what can become a viable product.
Selected work
Work in progress
-
-05ReferralInfrastructure for agentic web
- -04Design Cognition & Behavioral StudyHCI
- -03Agent CoreCross-channel AI product
- -02Threshold Design ProgramSocial infrastructure
- -01A Living Archive of Community Ownership ModelsArchive
Selected Projects
[2023 – 2026]
Hi, I’m Yvette.
Mostly mid-stride.
I let ideas grow wild, then trim them, cut by cut, into a frame that can be built.
I came to social science because I kept asking why people do what they do. I still haven’t stopped.
(an ongoing journey)I believe we learn by building, not by being told. So I trained myself to make things: when I want to understand something, I build a version of it.
(constructivism, applied to myself)After graduating, I spent a year gathering design skills.
(2024–2025)Then I stepped into design school, and slowly found what I keep coming back to: the systems underneath how people, and now machines, decide and act together.
(Harvard GSD, class of 2027)Give me a field I don’t know yet. I read everything, find the people inside it, and come back speaking their language in words everyone else can use.
(my favourite part of any project)Most of my work happens in rooms full of people who want different things. I am the one translating.
(+86 · +44 · +1 · +852)And when it gets loud, I find a window, sit in it, and think.
(then I go back and build)
What I do
Every skill I have, I needed first.
I chase the problem, not the toolkit.
I learned social science, theory and method, to take a structure apart and see why it holds together the way it does. I learned design, fabrication, visualization, methodology, to build something that could act on what I found. I learned AI and infrastructure because the problem in front of me needed it. None of it was the goal. The problem was.
Drop me into a problem outside my training, and I will go acquire whatever it takes to solve it: a research method, a fabrication process, an unfamiliar piece of software, a new part of the AI stack. I do not wait until I am already qualified to start.
Method and essays
Method:
[How I work / ongoing]
Hold complexity long enough for negotiation to become possible. Between private experience and public structure runs a seam, one that was designed, and is actively maintained. I design and build tools that make it legible, contestable, and deliberately crossable.
After three years tracing habitus, my thesis landed on cleft habitus: the condition of being caught between two mutually untranslatable worlds, forced to hold two logics of action at once. It is structural, produced by capital, institutions, language, and access to information, then internalized as the pattern by which we react.
I refuse to romanticize this position. Bilingual advantage and cross-cultural vision give me not an advantage but constant friction, and friction is a force. I convert that dislocation into method rather than curing it.
Across contexts, I care about how occlusion becomes a problem, and how it becomes an entry point.
Restricted information flow constrains the thinking that could produce change. Information silos make a system’s complexity available to only some. I can’t erase the inequalities of institution, information, and resource, so I design interfaces that at least make them visible, contestable, and acknowledgeable.
Surface the institutional grammar, decompose a seemingly natural outcome into deliberate moves, and hold any settlement accountable before and after it occurs.
The represented are never silent; their speech simply isn’t on the channel that gets heard. I build the channel, the surface, and the shared object.
I wrap evidence of how power relations and capital flows operate into formats different actors can read, question, and use.
- Mapping as understanding. Mapping is how I refuse to commit to a narrative too early.
- Most appropriate expression. Information is filtered, hierarchized, optimized for external readability.
- Social science awareness. I treat my own position as dynamic, relational, and self-aware.
- Perspective shifting. I build mechanisms that make readers inhabit several perspectives at once.
- Complex, not complicated. I compress the mess into a takeaway a smart non-expert can follow.
- Evidence without distortion. I phrase evidence in the least-distorting way that lets readers keep judgment.
- Permission to say I don’t know.
- Permission to show messy tooling beside polished artifacts.
- Permission to hold a position, as long as it is acknowledged.
- Permission to make something discussable before calling it solved.
Academic writing: UCL papers, 2021–24
In progress
.WAn agentic writing toolI love writing, so I am turning how I write into a product.