MetaVoice transforms phone-based customer service with Oracle AI infrastructure
MetaVoice trains its multimodal AI models 15X faster using OCI's GPU cluster and ultra-high bandwidth network for phone-based customer service.
“Oracle AI infrastructure provided a combination of superior training performance, expert support, and ability to scale to our growing community of business users.”
MetaVoice is on a mission to build human-grade, AI customer service representatives to automate business phone operations at the scale of a billion users. The technology company needed fast and reliable GPU compute clusters to train its real-time multimodal AI and wanted to move quickly without the burden of setting up and managing its own infrastructure. MetaVoice chose Oracle Cloud Infrastructure H100 GPU clusters and high-speed RDMA network between bare metal instances that were 50% more cost effective compared to other cloud providers. OCI’s RDMA network boosted performance dramatically enabling 3 Tbps communication across the cluster. As a result, MetaVoice was able to train models faster, reducing time to market. In addition, the company tripled its storage to 55 TB per node, thereby reducing complexity with streaming data over the network and paying heavy egress fees.