Artificial intelligence unicorn Poolside Inc. today announced it’s collaborating with the public cloud infrastructure giant Amazon Web Services Inc., making its foundational models and generative AI assistants available through the Amazon Bedrock service.

The move, announced at AWS re:Invent 2024, will enable customers to access Poolside’s foundation models privately and securely to create their own, customized coding assistants using their own data, the companies said.

Poolside last hit the headlines in October when it bagged $500 million in a Series B funding round led by Bain Capital Ventures, with participation from Nvidia Corp., eBay Inc. and many others. The startup has designed a series of generative AI coding models for writing software, and is a rival to Github Inc.’s popular Copilot. Its models can help developers with tasks such as auto-completion of code, or suggesting lines of code that might be relevant in a particular context.

According to Poolside, its models are already very popular, used by numerous Global 2000 enterprises, though it has only disclosed a handful of them.

Poolside’s models stand out thanks to their reinforcement learning-based approach to training, which allows them to learn over time based on code execution feedback. As such, its models gradually get better and better as they work. The startup also provides developers with a software toolkit for AI training tasks, tools for measuring the accuracy of its models and more.

Amazon Bedrock is a fully managed service that provides developers with access to high-performance foundation models from a number of leading AI companies via a single application programming interface. The advantage of bringing its models to Amazon Bedrock is that it makes it easier for organizations to satisfy their privacy, security and compliance requirements when using Poolside’s models. It allows customers to deploy the models in the same environment as the data they intend to train them with, within their security firewalls.

The collaboration will also see Poolside use Amazon’s AWS Trainium chips, which are specialized AI accelerators, to power inference for its Malibu and Point models, ensuring optimal price performance for customers. In other words, customers can now deploy Poolside’s models through Amazon Bedrock, using the Amazon Elastic Cloud as a data store, and run them on the AWS Trainium chips at lower costs.

This isn’t the first time the companies have worked together. Poolside’s Reinforcement Learning From Code Execution Feedback technique already runs on AWS’ infrastructure, which supports up to 10,000 code executions per minute. The RLCEF technique makes it possible for Poolside to generate synthetic training data at scale, which is done by exposing its models to realistic coding tasks on real-world codebases and providing feedback on the model’s work. By scaling RLCEF, Poolside says, it can continually improve the quality of its models without having to use its customers’ data to enhance them.

Poolside Chief Executive Jason Warner said customers will also benefit from streamlined procurement, contracting and budgeting when accessing its models on AWS, making it easier than ever to start experimenting with them.

Image: SiliconANGLE/Microsoft Designer