Product manager who builds.
I find where people wait, decide what to build and what to leave out, and ship the first version with AI agents. I lead product at Devtron. My work spans enterprise B2B, AI agents, health and internal tools.
| Who waited | For what | Before | After | Product |
|---|---|---|---|---|
| An AI team | a GPU | now self-serve | GPU Platform | |
| An engineer | a database query | now 17 min | BharatPe portal | |
| An on-call engineer | the cause of a failed deploy | now ~1 min | Atlas | |
| A parent with diabetes | a medicine reminder | a family call | In development | Sakhi |
Work
All case studies- Devtron · AI infrastructure
GPU Platform: GPUs without tickets
AI teams waited days for a GPU and weeks to ship a model. I defined a self-service GPU platform where infra teams set the rules once and AI teams move on their own.
80%+fewer GPU tickets - Devtron · AI assistant
Atlas: from hours to a minute
When a deployment failed, users read raw logs for hours or opened a support ticket. I shipped Atlas, an AI assistant that finds the root cause and suggests a fix.
was hoursnow minutesto recover from a failed deploy - Devtron · AI agents
AI Control Plane: agents you can trust in production
Companies had no safe way to deploy AI agents or control what they do. I defined one control plane to build, deploy and govern agents.
3design partners in private beta - Devtron · Growth and enterprise
Devtron Platform: a trial that converts, and approvals enterprises need
Trial users left before they saw value, and regulated enterprises could not adopt us without sign-off before production changes. I fixed both.
4xinbound leads - BharatPe · Internal developer platform
BharatPe: 14 days to 17 minutes
Engineers waited 14 days for a small DevOps team to run database queries and server changes. I built a self-service portal with access rules and audit logs.
20,000+requests through the portal
Building now
All projectsIndependent products I am building in my free time. None is live yet.
- Consumer health · Voice AI
Sakhi: an AI nurse that calls first
Older Indians with diabetes, BP and thyroid miss medicines and readings unless family reminds them. Sakhi is a voice-first AI nurse that calls them on WhatsApp in Hindi.
Doctor and legal review are open. - Clinical AI · Retrieval
Dhanvantari: answers a doctor can check
Indian doctors lack a trusted clinical Q&A tool that covers Indian guidelines and medicines. Dhanvantari answers clinical questions, and every claim links to its source.
The eval gate has not run on real answers yet. - AI agents · Reliability
Peira: AI agents that improve from their own failures
AI agents keep failing in production, and teams find and fix each failure by hand. Peira finds repeated failures in an agent's traces, proposes a fix, and proves it before a person approves it.
No customers yet.
How I work
The full methodDiscoverDefineBuildTestLearnback to Discover.
I find the real problem, write down what not to build, and build the first version myself to learn fast. Then I test it with real users and real data.
Notes
All notes- Why our health agent sounded scripted, and the tests that hid itEvery test passed, and the agent still sounded like a phone menu. The tests checked that it followed the script, not that it understood the person.
- Remove the wait, keep it safe: one idea in five productsA GPU platform, a debugging assistant, an ops portal, an AI agent control plane and a health agent look unrelated. They solve the same problem.
- Changing a large MySQL table without downtime, with gh-ostA plain ALTER TABLE on a large, busy MySQL table can block writes for minutes. This is how to run the change with gh-ost: test it on a replica, verify the data, then cut over safely.
- Automating Redash permissions with the Redash APIEngineers waited on an admin to get access to Redash query groups. A small Python client on the Redash API let an approval flow grant access in seconds, with an audit trail.