Replicate Inc., a startup making it easier for software teams to build artificial intelligence applications, today exited stealth mode and announced that it has secured $17.8 million in funding.

The company raised the capital over two rounds. It received the bulk of the funding, $12.5 million, through a Series A investment that included the participation of Andreessen Horowitz, Y Combinator, Sequoia Capital and multiple angel investors. It earlier closed a $5.3 million seed round.

Incorporating AI models into an application is often technically challenging. Neural networks usually can’t run on their own, but rather have to be deployed together with several auxiliary code components. Those components can be difficult to write and manage. 

Another source of complexity is the fact that AI-powered applications are often deployed on graphics cards. In some cases, adding graphics cards to an application environment can complicate infrastructure management tasks.

Replicate offers a platform that promises to reduce manual work for software teams. According to the startup, the platform enables developers to deploy AI models with a few lines of code.

“There are roughly two orders of magnitude more software engineers than there are machine learning engineers (~30 million vs. ~500,000),” Chief Executive Officer Firshman wrote in a blog post today. “By building good tools, we think it is possible for software engineers to use machine learning in the same way they can use normal software.”

As part of its feature set, Replicate offers a collection of several thousand open-source AI models. The AI models can perform a variety of tasks ranging from generating images to translating text. Its platform also works with custom neural networks that companies develop in-house. 

Neural networks are often deployed with auxiliary components such as PyTorch and TensorFlow, tools that help speed up AI development. In many cases, developers also add more specialized components such as Nvidia Corp.’s CUDA toolkit. The toolkit is used to optimize the performance of AI models deployed on Nvidia chips.

Before an AI model is deployed, developers must package the model and its supporting components into a software container. The process involves writing a code file known as a Dockerfile. The file defines what components the container should include as well as how they will be configured.

Replicate has developed an open-source tool called Cog that eases the task. The tool, which is included in the company’s platform, enables users to configure containers with less customization than the task usually requires. Cog can generate a Dockerfile, the files that define a container’s configuration, based on a limited number of instructions provided by software teams.

The platform also eases the task of deploying AI models in production. After developers package a neural network into a container, Replicate can automatically deploy it on a managed cloud environment. The environment automatically adds and removes hardware resources as graphics cards based on usage.

Replicate is gaining significant traction in the developer ecosystem. According to TechCrunch, the startup’s user base has been expanding at a 149% month-over-month growth rate since mid-2021. 

Photo: Unsplash