Head of Platform Engineering · builder of supervised AI agents
I build supervised AI agents for real operations: a vertical agent platform for healthcare at mymediset — in pilot with two enterprise healthcare customers — and Ops Room, an operating agent for e-commerce founders. I write about what makes agents dependable: tools, context, evals, permissions, and recovery paths.
Built and shipped
Vertical AI agents for medical-device consignment, surgical kits, and hospital procurement, in pilot with two enterprise healthcare customers. Earlier at mymediset: a mobile-first booking app that cut booking times by 50% (App Store · Google Play) and an AWS object-detection system that cut warehouse identification effort by 90%.
A supervised operating agent for e-commerce founders: approval-gated execution, connector integrations, durable state on Cloudflare Workers, and an eval harness with regression baselines and nightly live runs in CI. Live at oproom.ai.
clify and its generated CLI fleet run my own store's daily operations; weave is on npm; organizational-seed runs an organization as plain files. More at github.com/codeyogi911.
Latest writing
I am building Mainmind: a governed knowledge layer that gives people and AI agents the same account of how a company works, without trapping that knowledge inside any model or application.
We keep bolting rigid systems around agents. The better architecture is a firm organizational core with a disposable, evolving harness around it.
Agent experience is not an alternative to agent evals. It is the product surface those evals should help us diagnose and improve.
DevOps made shipping and operating software a continuous discipline. LoopOps is the next step: making improvement itself a first-class, measurable loop for humans and agents.
Operating principles
AI systems only become durable when quality, regressions, and risk are visible to the whole team.
Better tools, context, permissions, and recovery paths create more value than simply adding tokens.
Ship narrow, observable workflows first. Expand autonomy when the system has proved it deserves trust.