All Posts

A case study of Zenith, the AI support agent I designed and built for a live-events ticketing app. One model, a bounded tool loop, human approval for anything irreversible, and the ops tooling that makes an agent safe to change: a replay lab, a cost ledger and a dashboard.

Kuch karna hai? Chhotu se karwa lo. I built a personal AI assistant on Telegram that handles my GitHub, CleverTap, and DigitalOcean — so I don't have to. Here's the story of how a lazy engineer built himself a very obedient bot.

Most payment systems are built by teams of 20 or more. I built one alone — hardware, backend, dashboard, the whole thing. This is the honest story of what that actually took, and what I'd tell any solo engineer attempting something like it.