Case Study: Qwerky AI achieves faster cross-platform AI deployment with Modular

Modular allows Qwerky AI to do advanced AI research, to write optimized code and deploy across NVIDIA, AMD, and other types of silicon

Qwerky AI, a company building hardware-agnostic AI for conversational agents, faced significant barriers with traditional AI infrastructure. Frameworks like PyTorch were inefficient for their custom Mamba-based models, and writing platform-specific code for NVIDIA and AMD GPUs was complex and time-consuming. This fragmentation threatened their core mission of making AI accessible on any commodity hardware, leading them to seek a unified solution from Modular.

Modular provided its platform, utilizing the MAX graph API and the Mojo programming language, to solve Qwerky AI's challenges. This allowed Qwerky's small team to write optimized, hardware-agnostic kernels in a fraction of the code and deploy the same solution across different GPUs. The results were transformative: Qwerky AI achieved 50% faster GPU kernels, a 3x increase in research velocity, and the ability to deploy their novel architecture efficiently across their entire heterogeneous infrastructure.

Qwerky AI

Evan Owen

Chief Technology Officer