I'm Rajkumar Vallepu. I build AI agents that take over the repetitive part of operations work.
They follow your company's own policies, and a person on your team approves every reply before anything happens.
Hi Tom,
A refund is processed within 3 business days of the returned item arriving at the warehouse, and your return arrived 6 business days ago.
I am issuing your $180 refund now.
Thanks,
Larkfield Goods support
Caught by a check before it reached the customer.
The refund needed the team lead's approval, so the draft couldn't promise it yet.
Ticket T-002, from the agent's evaluation run. Play it back
What I'm building
I'm building an AI agent that helps operations teams with their ticket queue.
It reads incoming operations tickets and sorts them by type and urgency. When it needs more context, it looks up data in your own systems through tools, for example an order database or an internal docs page. Then it drafts the next action, such as a reply or a refund request. A person on your team reviews the draft and approves it before anything happens. The agent never acts on its own.
I'm keeping it configurable, so the same agent can work with any business's docs and systems. Each business is one profile file: its knowledge base, its categories, who handles what, its limits and its tools. You pick the profile, and the agent works with those.
The proof
On 15 tickets it had never seen, the agent got 94% of its decisions right: 66 of 70 fields, on its first run. Every one of those drafts still waited for a person to approve it.
| Business | Fields right | Tickets fully right |
|---|---|---|
| Online store | 46/50 | 8/10 |
| Property management | 20/20 | 5/5 |
| Both | 66/70 | 13/15 |
Unseen tickets, written with their expected answers before any model run. First run, 10 October 2026, on gemini-3.5-flash-lite.
Two made-up businesses, an online store and a property manager, set up with no code changes.
ops-agent, an AI operations triage agent
My current project. It sorts support tickets, checks orders and payments, applies refund limits, and drafts replies for a reviewer.
Experience
I've been building Java and Spring Boot systems for over five years, mostly in payments and enterprise software, where a wrong number isn't a small bug. That's where my habits come from: tests before a change goes in, limits enforced in code, and an audit trail for every decision.
- Built an AI assistant for an operations team; the team estimates it cut investigation time by about 40%.
Java, Spring Boot, Spring AI, Postgres and pgvector, REST APIs, React and TypeScript, Docker, payments systems.
Contact
If your team answers the same questions all day, I'd be glad to show you how this would work with your own policies. Please do reach out.