Backflip Inc., a startup helping engineers create 3D models faster, today announced that it has closed a $30 million early-stage funding round.

NEA and Andreessen Horowitz led the Series A raise. They were joined by a Who’s Who of angel investors: Microsoft Corp. Chief Technology Officer Kevin Scott, Android founder Rich Miner and Ashish Vaswani, a co-inventor of the Transformer neural network architecture. Backflip previously raised a seed round of undisclosed size. 

Backflip was founded in 2022 by Chief Technology Officer David Benhaim (pictured, left) and Chief Executive Officer Greg Mark (right). The duo previously launched Markforged Inc., a major supplier of 3D printers. The latter company was taken private for $115 million earlier this year. 

Before engineers can make an object using a 3D printer, they have to supply the printer with a blueprint. This blueprint takes the form of a 3D model created with the help of computer-aided design, or CAD, software. Designing such models can take days in some cases.

One reason the task is time-consuming has to do with the way 3D models are implemented. 

At its core, a 3D model is a point cloud, a collection of points that forms an outline of an object. Engineers fill the gaps between those points by adding in various geometric shapes using their CAD tool. When designing complex objects such as an engine part, the process can involve a significant amount of manual work.

Some 3D models, such as the blueprints of objects that must be water-proof, present additional challenges. 

When engineers draw a 3D model manually in a CAD application, there’s a risk that they will accidentally leave empty spaces between some of the object’s pixels. During manufacturing, the 3D printer will turn those spaces into gaps in the object’s structure. Engineers avoid such gaps using a technique called constructive solid geometry, which involves assembling 3D models from existing building blocks rather than drawing them from scratch. 

Backflip is working to speed up the 3D modeling process. It has developed an artificial intelligence platform that can automatically generate 3D models based on a text description provided by the user. According to the company, the software can also take a photo or a sketch as input.

Under the hood, Backflip’s platform is powered by an AI system developed specifically for 3D modeling tasks. The algorithm features a “novel neural representation that teaches AI to think in 3D.” According to the company, AI models based on its technology provide better resolution other algorithms and require less hardware to run.

“That development yields 60x more efficient training, 10x faster inference and 100x the spatial resolution of existing state of the art methods,” Benhaim said.

According to TechCrunch, Backflip trained its AI using a dataset of about 10 million 3D part blueprints. Some of the data is synthetic, meaning it was generated by a neural network for AI training purposes.

Backflip says that its platform reduces the amount of time required to create 3D models from days to a few minutes. Blueprints created by the software work with 3D printers that use metal, carbon fiber and plastic for manufacturing.

Photo: Backflip