logo for Rendered Ai and photos of planes on runways
An arrow pointing leftHome

Physics-based Rendered.AI gets $6 million in seed funding


The Bellevue-based startup is creating synthetic datasets for training AI systems

Let’s face it: Data sets are pesky things. Critics will say they are inherently biased; even data scientists who have drunk the Kool-Aid admit they spend most of their time cleaning them up before they can use them to train their AI. And some emerging sectors don’t even have data sets to begin with.

Which is why physicist Nathan Kundtz, who in his former life was the CEO of Bill Gates’ satellite communications company, Kymeta, set out to create synthetic data sets that also allow engineers to customize them. This simple yet brilliant idea just landed his new startup, Rendered.AI, $6 million in a seed round led by Space Capital followed by Tectonic Ventures, Union Labs, Congruent Ventures, and Uncorrelated Ventures. The company is presently in beta mode, but already can provide training data sets for medical, automotive, security, X-ray, and robotics applications, as well as for satellite communications companies. (Obviously — Kymeta raised $440 million, half of it under Kundtz’ leadership.)

Rendered’s two-year-old platform works by using industry-standard models and techniques that can be mixed and matched to create an unlimited variety of synthetic imagery for AI training optimization. Its cloud-native platform can scale quickly and allows data scientists to iterate on the fly for their needs. Kundtz said that he doesn’t believe synthetic data will replace real-world data, but that it can fill the gap for artificial intelligence applications, especially in nascent industries with AI/ML applications.

The Bellevue-based firm seems to be on a roll. Along with the funding announcement, the physics-minded synthetic data set company told attendees at the second day of the NVIDIA GTC conference that Rendered.AI’s Synthetic Aperture Radar (SAR) synthetic data simulation saw a 200-fold improvement in electromagnetic simulation calculation time with NVIDIA GPUs when pitted against the Intel CORE CPU i5-103OOH and a TESLA V100 -SXM2. Not bad for a company that’s barely a toddler.

“These improvements take us a long way to being quite speedy with our calculations,” said the CEO.