Hippocratic AI, a company that builds patient-facing AI agents for non-diagnostic healthcare tasks, has reached 10 million patient calls with its Polaris system, which runs on the AI infrastructure of DigitalOcean, a cloud computing provider serving more than 650,000 customers. The system operates on NVIDIA HGX B300 GPUs, the chipmaker's hardware for AI workloads, and reported a 99.9% clinical safety score across those calls.
The scale matters because of what Polaris does. Hippocratic AI builds AI agents that handle patient-facing clinical work such as chronic disease management, medication adherence and clinical scheduling, while not allowing its agents to prescribe or diagnose. Across more than 180 million patient interactions to date, the company reported an average patient rating of 8.95 out of 10 and said its safety evaluation involved more than 7,500 clinical staff. In this kind of deployment, an interruption during a call is a clinical event rather than a technical inconvenience, which raises the reliability requirements on the underlying infrastructure.
Over the past year, DigitalOcean and Hippocratic AI worked to optimise the inference stack, the layer of software and hardware that runs an AI model in production, for healthcare workloads. DigitalOcean engineered its platform with hardware-aware scheduling, optimised inference runtimes and scaling tuned for sustained high-concurrency use. Hippocratic AI contributed its own techniques, including FP8 and NVFP4 quantization, methods that reduce the computational cost of running a model, along with caching optimisations for long clinical conversations. NVIDIA provided early access to its B300 hardware and engineering support across its Hopper and Blackwell chip architectures.
The companies reported that, on long clinical sessions, the combination produced roughly 30% higher throughput per node and halved prefill latency compared with a prior-generation configuration. These figures build on results Hippocratic AI reported earlier in the month, when it cited a doubling of production inference throughput and a 40% reduction in end-to-end P99 latency on the same cloud.
"Polaris is built for the realities of clinical care: long sessions, real human conversations, zero room for error. With DigitalOcean and NVIDIA, we have early access to NVIDIA HGX™ B300 and the optimization techniques it unlocks, including NVFP4 quantization," said Debajyoti Datta, Co-Founder, Hippocratic AI. "That is what allows us to hold a 400-millisecond time-to-first-token at production scale, on the clinical conversations our patients depend on."
"What Hippocratic AI has built in healthcare AI is remarkable, hundreds of millions of real patient interactions across some of the most complex and sensitive moments in people's lives," said Paddy Srinivasan, Chief Executive Officer, DigitalOcean. "Delivering that at 99.9% clinical safety is what production AI looks like when it matters most. This is what purpose-built inference delivers, and it's what our AI-Native Cloud makes possible. Hippocratic AI's results are the proof."
"The demands of safety-critical AI workloads are fundamentally different from consumer applications," said Dave Salvator, Director of Accelerated Computing Products, NVIDIA. "DigitalOcean and Hippocratic AI are demonstrating how tightly integrated infrastructure and inference optimization, built on NVIDIA Hopper and Blackwell architecture, can deliver both performance and reliability at scale."
