Nuraj Kumaratilake

Product Manager · Co-founder

I’m a Product Manager and co-founder passionate about using AI to solve complex problems. I’ve led commercial product growth at REA Group and built and launched AI-assisted tools at The Arch Company.

That experience showed me how hard it is to piece together how a process works—and led me to co-found an AI venture focused on business discovery.

My path so far

Business analysis Product management Transformation Co-founded an AI venture for business discovery Today Business analysis Product management Transformation Co-founded anAI venture for businessdiscovery Today

A workspace for business discovery

Design prototypes · Working product in development

What we’re building

We’re building a workspace to bring conversations, documents and incomplete information together into a developing understanding of a process.

Understanding workflows through natural conversation

Luma is designed to interview employees naturally through voice or text to understand how their processes and workflows work. Prepared questions and practitioner guidance shape the conversation.

  1. Prepare the interviewMaya sets the questions and invites Sam.
  2. Conduct interviewLuma interviews Sam to understand how the process works.
  3. Uncover an exceptionLuma follows up on what Sam just said.
  4. Check and correctSam adds missing detail before sharing.
  5. Share with MayaMaya receives the contribution and correction.

A document that develops with the discovery

DecisionMake the developing document the place where the work happens, not mainly an output.
WhyDiscovery is iterative: people start with fragments of information and build understanding over time. They need to move between incomplete areas and add detail as they learn.
What I keptEditing, linked questions and sources from September. What changed is how the work happens.
SeptemberIndividual steps, with the process document in a separate view.
LatestWorking directly in the developing document, adding detail as understanding grows.

Reviewing AI changes where they happen

DecisionReview an AI change inside the step it affects, and keep fuller review for larger or conflicting updates.

The earlier design showed proposed changes and their sources on a dedicated review screen. The newer design also lets someone ask Luma for a correction and review the proposed change directly in the relevant process step.

I wanted AI assistance to fit into the work without adding a long queue of approvals.

SeptemberProposed changes and their sources were inspected on a dedicated review screen.
LatestAsk Luma for a correction, then edit, accept or reject it inside the step.

What I learned testing it

“Discovery is iterative. Each conversation should build on what’s already known.”

I built a multi-agent workflow for internal product evaluations, combining UX, QA, business analysis and simulated stakeholder roles. Testing exposed gaps in how the product handled incomplete and conflicting information across discovery sessions.

This led me to redesign the workspace around a developing understanding of the process, so people could reconcile inputs, revisit gaps and build on what they’d learned over time.

More to come...