Hippocratic AI's Polaris system has processed 10 million patient calls while maintaining a 99.9% clinical safety score, running on DigitalOcean's (NYSE: DOCN) AI-Native Cloud platform powered by NVIDIA HGX B300 GPUs, according to a company announcement.

The healthcare AI system has handled more than 180 million patient interactions across chronic disease management, medication adherence, care gap closure, and clinical scheduling. Human evaluation involving more than 7,500 clinical staff supported the safety assessment, with patients rating the service an average of 8.95 out of 10.

DigitalOcean's platform delivered performance improvements including 2× prefill speedup and approximately 30% higher per-node throughput compared to prior-generation configurations. The optimization work involved collaboration between DigitalOcean, Hippocratic AI, and NVIDIA teams over the past year.

"Polaris is built for the realities of clinical care: long sessions, real human conversations, zero room for error," said Debajyoti Datta, Co-Founder of Hippocratic AI. "That is what allows us to hold a 400-millisecond time-to-first-token at production scale."

The technical improvements included hardware-aware scheduling, optimized inference runtimes, FP8 and NVFP4 quantization, KV-cache optimization, and custom MoE kernels. NVIDIA provided early access to HGX B300 hardware alongside engineering collaboration.

Hippocratic AI has raised $404 million in funding from investors including Andreessen Horowitz, General Catalyst, Kleiner Perkins, and NVIDIA's NVentures. The company focuses on non-diagnostic patient-facing clinical AI agents and prohibits its technology from being used for prescribing or diagnosing.