Skip to main content
Back to Blog
InsightsProduct3 min read-

What we decided in Atami: one platform, built from Japan

In May, the Ur AI team spent two days together in Atami, families included. We agreed to bring Nebula and Specter together as one platform, built from Japan, starting with document-heavy professional work.

Sandeep Yella

Sandeep Yella

Founder & CEO

What we decided in Atami: one platform, built from Japan

In May, the Ur AI team spent two days together in Atami, families included. Incubate Fund sponsored the stay as part of their support for the startups they back. We started the first morning with a small pooja led by my Mom, and the Gayatri Mantra set the tone: ask for a clear mind, and let the work speak.

We started with people

Before talking strategy, each of us shared what we want from the coming year, at work and outside it. Learning, growth, family, a big personal challenge, a chance to travel. We built our company plan on top of those answers.

Sandeep Yella beside a whiteboard listing the goals the team shared for the coming year
Writing up what each of us wants from the year, at work and outside it.

Nebula and Specter as one platform

Our biggest decision was how to describe what we build. In Atami we agreed to bring Nebula and Specter together as one platform.

Nebula is the foundation: it turns complex business documents into data that AI can work with accurately and securely. Specter, our AI for M&A due diligence, is the first module on top of it. More modules will follow for other document-heavy professions, such as for construction or real estate or insurance.

Our aspiration is simple to say and hard to do. Today every enterprise runs on a cloud such as AWS. In the AI era, every company will run its own models on its own data, and we want to be the layer they build on. We want to be the AWS of the AI era.

Japan is the right place to start

Japan still runs on paper more than most countries. Cloud adoption is lower than in the US, and the bar for accuracy and compliance is higher. Many companies want AI that runs on their own infrastructure.

At the same time, the cost of using frontier models is becoming a real worry for businesses. We believe the winning offer combines four things: security, predictable cost, sovereignty and accuracy. If we can deliver that in Japan, we can deliver it anywhere.

Team members and family getting plates ready at a charcoal barbecue in Atami
Barbecue on the terrace.

Two ideas we brought home

The first: the best founders define a clear first market and then keep widening it. We are starting with document-heavy professional work in Japan, and we intend to keep adding to that.

The second is an analogy I keep coming back to.

The frontier labs are building Superman. What most of our customers need is a reliable hero for the work in front of them, every day, at a cost they can predict.

What has happened since May

  • We completed our SOC 2 Type 2 examination and our GDPR compliance program, adding to the ISO/IEC 27001 certification we earned in January.
  • Nebula now comes in two versions: Nebula Frontier in the cloud and Nebula Sovereign, a fine-tuned model that runs fully on a customer's own infrastructure.
  • In July we published our benchmark of 14 document AI systems on Japanese business documents.
  • A veteran OCR researcher joined us to lead research. More on that soon.

Later this year we plan to visit San Francisco, Shanghai, Bangalore and Mumbai to learn from the teams building AI there.

Team members and family playing a board game around a low table
Board games with the families.

Grateful, and excited

Atami reminded me why we do this. A small team from many countries chose to build something hard together, from Japan. Thank you to everyone who came, and to Incubate Fund for making it possible. The road ahead is long, and we are ready for it.