hi, i'm shreyan.

i study how multi-agent llm systems fail, mostly by attacking them (my paper when the judge intervenes was accepted at a neurips 2026 workshop), build products (i made ice, a consumer social app, and have worked in product at two early-stage startups), and explore what interfaces for thinking with ai should look like (i recently built essay atlas, an llm-drawn map of where my essays connect and contradict). i also write essays about design, media theory, and ai on my substack.