About

Hi, I'm Vivek Verma— a full-stack engineer from Haridwar, India. I've shipped 0-to-1 systems with real users behind them: a real-estate platform now serving around 30K daily visitors, an offline-first point-of-sale system running across 11 terminals, and a nine-month enterprise PLM build I led end to end with direct client ownership.

I like building real-time, offline-first products that hold up when the network doesn't — push notifications that ring through a killed app, sync layers that survive a 2G connection, idempotent delivery so a retry never doubles a message. The part I enjoy most is a layer lower: I once traced multi-minute reports in a 15-year-old database down to missing keys and absent indexes, and got them under three seconds.

That instinct is what pulled me into LLM inference— the same questions, harder constraints. What does a token actually cost, where does the latency go, and at what point does an optimisation stop paying for itself? I'm working through it from first principles and publishing every measurement, including the ones that didn't work, in my daily log.

Find more on GitHub, LinkedIn, or X.