About

Why I built DwaraOne

One door. Every HR journey. Governed on Databricks.

It's 11 p.m. in Singapore. A cabin crew member wants to know one thing: can she take leave before Saturday's flight to London? The answer lives in four systems. The leave ledger. The training platform. The roster. The policy library. Tonight she'll open none of them. She'll ask a colleague, get it half right, and find out the hard way.

Nobody quits over one form. They quit over forty. Over portals that don't talk to each other. Over asking a simple human question and receiving a ticket number. HR isn't hard because people are complicated. It's hard because the answers are scattered and the actions are buried.

DwaraOne is my answer. Dwara means doorway. One door an employee walks through for everything: ask a question, take an action, track the result. No forms. No portal-hopping. A conversation.

The shift from Ask Me Anything to Task Me Anything is no longer a future vision. It is here today. Agent Brahma is Task Me Anything for People Ops.

Behind the door is an agent, not a search box. Agent Brahma reads your live record before you finish typing. He notices your safety recertification is due before your next roster and tells you before you ask. He knows your six years of tenure means 24 leave days, not 22, because he joins the policy document with your employee record at the moment you ask. He books your leave, adds your newborn to coverage, and requests your software, all in one conversation. And when a question needs a human, he says so and routes it. Knowing its limits is part of what makes an agent trustworthy.

The trust is the architecture. Every answer is grounded in governed gold views on Databricks, and every answer shows its source. Every transaction runs one lifecycle: preview, confirm, submit, approve, backend update, audit. Nothing touches your record without your confirmation and an approval trail. Writes land in Lakebase, Databricks' transactional Postgres, where every state change is appended to a hash-chained audit log, each entry cryptographically linked to the one before it, then merged into Delta tables under Unity Catalog. The same governance that protects the company's analytics protects the employee's paycheck. I've spent my career in enterprise data, from SAP HANA migrations to lakehouse platforms, and this is the lesson it taught me: an agent is only as trustworthy as the data substrate underneath it.

AI needs context. Agents need governance.

The system also learns. When a policy confuses people, that confusion is captured as a signal, and leadership sees which policies to simplify. Employee friction literally becomes the improvement backlog. Managers get the same agent with a different lens: risks first, approvals in chat, one place to decide.

The ideas underneath this are published. The preflight checks are a working world model: the agent simulates an action's consequences against the roster, the policy, and the ledger before anything executes. The life-event journeys are hierarchical planning: one human sentence decomposed into governed transactions. Both build on my research on data engineering-driven world models and hierarchical planning for consequence-aware agentic systems.

This is what data plus AI makes possible now. Not a smarter chatbot bolted onto old portals, but a different shape of workplace software: proactive instead of reactive, conversational instead of navigational, governed instead of merely convenient. The technology to end employee friction exists today. What's been missing is the will to build the door.

What's next: voice as a first-class interface, so crew can ask between flights, hands free. More journeys, from onboarding to relocation to retirement. And beyond HR, because every enterprise function, IT, finance, facilities, deserves its own front door on the same governed substrate.

And I'm open-sourcing this. The patterns here, the agent tool design, the preview-confirm lifecycle, the hash-chained audit, the Lakebase-to-Delta write path, shouldn't live in one demo. They should be a starting point for anyone building governed agents on the lakehouse. Take it, build your own door.

Every employee deserves to spend their energy on their work and their life, not on the systems in between. That's the whole idea. One door. Walk through.

— Brahma Reddy
Built on Databricks · Unity Catalog · Lakebase · Delta · Genie