Founder’s Notes
On automation, and where it should stop
Technical write-ups and short essays from building InnoveBots. Mostly about where software should stop.
A Supervision-First Agent Architecture
Our architecture for supervised LLM agents: how we constrain what an agent can do, and why the cost of error in a small business is an architectural problem rather than a tuning one. Evaluation protocol included; results to follow after the private beta.
More notes are in progress. If you’d like them by email when they go up, join the waitlist.
Yatendra Kaushik
I studied at IIT Ropar and now build InnoveBots from Dubai. It is a workflow automation with AI agents on top, for businesses operating in the UAE. We closed a pre-seed round of AED 1.5 million and are currently in private beta.
Most of what I write here comes out of that work: where automation helps, where it shouldn't reach, and what I've got wrong along the way.