How financial misinformation spreads across Indian social media, under the mentorship of a professor at the Indian School of Business.
The content is multilingual and code-mixed, Hindi and English inside the same sentence and sometimes the same word, which is exactly what most off-the-shelf language models fall over on. So the pipeline runs on MuRIL and XLM-RoBERTa instead.
The other half is psychological rather than computational: the BFI-2-XS personality inventory and the Fear of Missing Out scale, to work out which individual traits predict whether someone forwards a claim they never checked. The interesting question is not what the false posts look like. It is who passes them on.
Running the Edunomix India Chapter, which I founded to teach economics to students who never get offered it. Most of my week goes here: curriculum, chapter leads, and the unglamorous work of keeping volunteers energised.
Also serving as CFO at TaxCity, where I spend more time in spreadsheets than I ever expected to enjoy.
Most of my writing now goes into the study above, which is a different craft from an essay. Fewer adjectives, more error bars, and a much longer wait before you are allowed to claim anything.
Shorter pieces land on the articles page when they are ready. The Two Faces of Progress is done and published, and I have stopped picking at it.
New episodes of Decoded, my podcast on economics and technology. The hardest part was never the recording. It is deciding which of my own takes to cut.
- Principles for Dealing with the Changing World Order Ray Dalio
- The Policy Pivot Ajay Khanna & Rahul Sharma
- A Song of Ice and Fire George R. R. Martin · for when policy gets heavy
The whole shelf lives here.
Why a false claim about a stock travels faster than a true one, and whether that is a property of the claim or of the person holding the phone. I suspect it is mostly the second, which is the uncomfortable half, because you can moderate a post and you cannot moderate a personality.
Whether AI governance can be written fast enough to matter. Every draft I read is careful, thorough, and about eighteen months behind the thing it regulates.
Also: why economics is taught as a subject to be examined rather than a lens for seeing the world. That gap is most of why Edunomix exists.
To be technical. I came to this from economics, where it is easy to treat the modelling as somebody else's department. That stops being an option the moment your research question is a fine-tune. So: Python, the transformers stack, and enough statistics to know when I am fooling myself.
Number theory, still, slowly. The apprenticeship ended but the habit did not.
Claude Code. I have been handing it the parts of the job that are syntax and keeping the parts that are intent, which is a strange and slightly unnerving way to work. Most of this site is downstream of that habit. I still cannot decide whether it has made me faster or merely braver.
Most things. It took me a while to learn that a calendar with no gaps in it is not a sign of ambition, it is a sign of poor editing.