Most of these questions that shape my research interests didn't arrive fully formed per se. They accumulated through building systems and watching them meet reality — an emotion recognition tool that kept running into the edges of what NLP could actually read, a financial literacy chatbot that surfaced how little our frameworks mapped onto the people we were designing for, a startup that raised the question of whether what the product measured had anything to do with whether it was actually helping. The tldr: the technical problem is usually tractable but what's underneath it tends not to be.
Being from Mauritius, growing up in Singapore, navigating three languages that carry different assumptions about how the world works — none of that makes me the most qualified person to answer what follows. But it somehow shapes what I notice. How you ask a question depends on which world you're standing in.
Currently working in trade and economic policy, somewhere at the edge of where data and diplomacy meet. The question that keeps surfacing: when a system fails someone (particularly the someone it was never designed for) what recourse do they actually have? I'm not sure the answer is a better tool.
When a tool does the inferring, whose assumptions does it carry?
Most conversations about AI flatten into productivity metrics. But tools reshape cognition, not just output — and the assumptions baked into how a system infers are rarely made visible. The OTR Copilot project taught me that the NLP was the easy part; the harder problem was whose emotional expressions the system was built to read, and whose it wasn't. A follow-on research question currently in development: do LLMs exhibit recognisable philosophical priors in therapeutic contexts — defaulting toward individual agency, reframing, resolution — and what does that mean for who they actually serve?
OTR Copilot → · AMIA 2026 → · UCL BCC → · Ethics of Anthropomorphic AI →
What happens when the framework for understanding a problem is itself part of the problem?
Economics gets taught as though its frameworks are neutral. They aren't — they encode assumptions about what counts as value, who gets to accumulate it, and what the state's role should be. The Reach Alliance research made this concrete: we started trying to build a financial literacy chatbot and ended up learning that the workers we were designing for already had sophisticated systems for managing money across borders. They needed structural support, not our frameworks. That gap between what a tool assumes and what a situation actually requires keeps showing up.
Reach Alliance case study → · What I learned → · I Was Wrong About Economics →
What does it cost to be the person the system doesn't have a box for?
Belonging is treated as a given — something you either have or find. But for people who grow up between cultures, languages, and systems of meaning, it's more like a practice than a destination. Singapore is an interesting place to sit with this: a cosmopolitan city that is also a very legible racial taxonomy, where you can be surrounded by people and still be categorically alone. Kura was five years of trying to build something for that space — $50K raised, 2,500 users, pilots with NUS and Duke-NUS. Closing it taught me things shipping it didn't.
The Experience of Being 'Othered' in Singapore's CMIO Classification →