• 01

  Latency budgets

  SSR caches that should render in 200ms, not 900. Lighthouse 95+ floor on every page I put in production.

• 02

  Voice for the people my parents’ age

  SwasthParivar understands how my father actually speaks Hindi — colloquial, regional, imprecise. Not textbook input.

• 03

  Agents that do work

  Not chatbots wearing a fancy name. The hard parts are tool use, schema constraints, and confidence-aware routing.

// Best fit: scaling a Next.js product, fighting a latency budget, or pulling a dashboard back from “the client bundle is 4MB.”