摘要：Hippocratic AI 旨在借助 AI 代理解决医疗行业临床人员短缺问题，通过推出支持语音功能的患者互动平台及创新 AI 代理应用商店，让临床医生无需编码即可开发、验证和分享适用于多种医疗场景的 AI 驱动脚本（所有脚本均需经过严格安全测试），进而扩大医疗服务覆盖面、提升护理质量。 为实现这一愿景，该平台核心聚焦面向患者的实时 AI 推理，这一需求对计算能力提出了极高要求，也带来了基础设施扩展性和资源分配方面的挑战。为此，Hippocratic AI 与 NVIDIA 展开合作，同时联合 AWS 进行部署，采用 NVIDIA H200 GPU、TensorRT-LLM 等产品及数据中心 / 云服务，优化推理性能以保障语音互动的无缝低延迟。
Hippocratic AI aims to combat clinician shortages in healthcare by harnessing the power of AI agents, which augments clinical staff with extra eyes, ears and a voice. To support this vision, Hippocratic AI has launched a voice-enabled patient engagement platform and groundbreaking app store where clinicians can develop, validate, and share AI-driven scripts tailored to a wide range of healthcare use cases. Each script undergoes rigorous safety testing before becoming available, and contributing clinicians help create safety-tested scripts that expand access and improve the quality of care for patients everywhere.希波克拉底人工智能（Hippocratic AI）旨在通过利用人工智能智能体的力量来解决医疗领域临床医生短缺的问题，这些智能体为临床工作人员提供了额外的眼睛、耳朵和声音。为了支持这一愿景，希波克拉底人工智能推出了一个支持语音功能的患者参与平台和一个开创性的应用商店，临床医生可以在其中开发、验证和分享为各种医疗用例量身定制的人工智能驱动脚本。每个脚本在投入使用前都经过严格的安全测试，参与的临床医生助力打造经过安全测试的脚本，从而扩大医疗服务的可及性，并提高全球患者的护理质量。
To achieve this vision, a core focus of the platform is to utilize patient-facing real-time AI inference. This capability demands substantial computing power, including the use of NVIDIA H200 GPUs, and presents challenges in infrastructure scalability and resource allocation. To address these challenges and ensure seamless, low-latency voice interactions, Hippocratic AI is collaborating with NVIDIA to optimize inference performance for on-demand deployment.为了实现这一愿景，该平台的核心重点是利用面向患者的实时人工智能推理。这种能力需要强大的计算能力，包括使用NVIDIA H200 GPU，同时在基础设施扩展性和资源分配方面也带来了挑战。为了应对这些挑战并确保无缝、低延迟的语音交互，Hippocratic AI正与NVIDIA合作，优化推理性能以满足按需部署的需求。
Hippocratic AI 希波克拉底人工智能
AWS 亚马逊云科技
Customized Inference 定制化推理
NVIDIA Hopper 英伟达霍珀
NVIDIA TensorRT-LLM
NVIDIA Data Center / Cloud NVIDIA 数据中心/云
Impact + Product 影响 + 产品
Impact + Product 影响 + 产品
The current healthcare system is under pressure with aging patient populations, aging workforces, and increased needs to do more with less. This is evident in practices like triage and risk stratification, which focus on the most urgent cases but can leave some patients at risk of falling through the cracks when it comes to future care needs.当前的医疗体系正面临压力，原因包括患者群体老龄化、劳动力老龄化，以及在资源减少的情况下需要完成更多工作的需求不断增加。这一点在分诊和风险分层等做法中显而易见，这些做法侧重于最紧急的病例，但在未来的护理需求方面，可能会让一些患者面临被忽视的风险。
In collaboration with NVIDIA, Hippocratic AI developed Empathy Inference, a technology for swift, natural conversations that forge emotional connections with patients.与NVIDIA合作，希波克拉底人工智能（Hippocratic AI）开发了共情推理技术，这是一种能实现快速、自然对话并与患者建立情感联系的技术。
AI agents can provide an infinite supply of care, speaking every language, remembering every conversation, and being clinically safe. This can help in providing continuous and comprehensive care to everyone, addressing the limitations of human resources. For example, AI can monitor patients during heat waves, check blood pressure daily, and provide timely interventions.人工智能智能体可以提供源源不断的关怀，它们会说各种语言，能记住每一次对话，并且在临床方面是安全的。这有助于为每个人提供持续且全面的护理，解决人力资源方面的局限性。例如，人工智能可以在热浪期间监测患者，每天检查血压，并提供及时的干预措施。
Hippocratic AI developed specialized generative AI healthcare agents—built on NVIDIA technology and deployed on AWS—to help shape the future of patient care. Expanding upon this, Hippocratic AI has launched its AI Agent App Store, a groundbreaking platform that empowers clinicians to design and deploy customized AI healthcare agents—no coding required. In just under 30 minutes, healthcare professionals can create agents tailored to specific medical tasks and workflows. Hippocratic AI 开发了专门的生成式人工智能医疗智能体——这些智能体基于 NVIDIA 技术构建并部署在 AWS 上，旨在助力塑造患者护理的未来。在此基础上，Hippocratic AI 推出了其人工智能智能体应用商店，这是一个开创性平台，能让临床医生无需编程即可设计和部署定制化的人工智能医疗智能体。医疗专业人员仅需不到 30 分钟，就能创建出适合特定医疗任务和工作流程的智能体。
The model is focused on learning from the healthcare experts—clinicians can create and contribute AI scripts for various healthcare use cases. These scripts are validated and safety-tested before being made available, giving patients faster access to safe, reliable care. Clinicians benefit by sharing their expertise at scale and helping to accelerate the development of new use cases.该模型专注于向医疗专家学习——临床医生可以为各种医疗用例创建并贡献人工智能脚本。这些脚本在投入使用前会经过验证和安全测试，让患者能更快地获得安全、可靠的医疗服务。临床医生则通过大规模分享自己的专业知识而获益，同时也助力加速新用例的开发。
The App Store debuts with over 300 AI agents spanning 25 medical specialties, supporting use cases such as cervical cancer check-ins, postpartum mental health monitoring, wound care, and diabetes screening. Every AI agent undergoes a rigorous three-step validation process, including licensure verification, development testing, and clinical review by a network of over 6,000 nurses and 300 physicians to ensure safety and efficacy.应用商店首次上线了涵盖25个医学专科的300多个人工智能智能体，支持宫颈癌检查、产后心理健康监测、伤口护理和糖尿病筛查等使用场景。每个人工智能智能体都经过严格的三步验证流程，包括执照核查、开发测试以及由6000多名护士和300多名医生组成的网络进行的临床评审，以确保其安全性和有效性。
“We trained our model differently than others and have always focused on inference, since our runtime is the key environment. Our model is actually 22 models—one gigantic 400B model doing the talking, with 21 others supervising it to ensure it doesn’t say anything unsafe. That’s a lot of inference—22 models’ worth. We’re running 4.2 trillion parameters each time, so we use up a ton.”“我们训练模型的方式与其他公司不同，并且始终专注于推理，因为我们的运行时是关键环境。我们的模型实际上是22个模型——一个庞大的4000亿参数模型负责生成内容，另外21个模型对其进行监督，以确保它不会说出任何不安全的内容。这涉及大量的推理工作——相当于22个模型的工作量。我们每次要运行4.2万亿个参数，所以消耗很大。”
Munjal Shah 蒙贾尔·沙阿
Co-Founder and CEO of Hippocratic AIHippocratic AI联合创始人兼首席执行官
Hippocratic AI is redefining patient engagement by prioritizing safety, accuracy, and empathy in its AI-powered clinical assistants. At the core is its Polaris constellation architecture, which deploys over 25 task-specific large language models (LLMs), each with 70B+ parameters, to reduce hallucinations and ensure clinical safety. These models—totaling over a trillion parameters—run on NVIDIA H200 Tensor Core GPUs using Amazon SageMaker HyperPod, delivering ultra-low latency and deeply empathetic conversations.Hippocratic AI通过在其人工智能驱动的临床助手中标注安全、准确性和同理心，正在重新定义患者参与方式。其核心是Polaris星座架构，该架构部署了25个以上特定任务的大型语言模型（LLMs），每个模型都有700亿以上的参数，以减少幻觉并确保临床安全。这些总计超过一万亿参数的模型在NVIDIA H200 Tensor Core GPU上运行，使用Amazon SageMaker HyperPod，可实现超低延迟和极具同理心的对话。
To enhance the natural flow of dialogue, the system accurately detects when a patient finishes speaking and responds without interruption—critical for building trust in clinical settings. Hippocratic AI leverages TensorRT-LLM to optimize its Polaris constellation of models, making them faster, smaller, and more efficient, which lowers costs and enables more conversations to run on the same hardware.为了增强对话的自然流畅度，该系统能准确检测患者何时结束发言，并在不打断的情况下做出回应——这对于在临床环境中建立信任至关重要。Hippocratic AI 利用 TensorRT-LLM 对其 Polaris 模型群进行优化，使其更快、更小、更高效，从而降低成本，并能在相同硬件上运行更多对话。
Given the sensitivity of healthcare data and strict HIPAA compliance, Hippocratic AI uses a robust multi-account, multi-cluster AWS strategy that separates production workloads from development environments. This secure, scalable infrastructure enables thousands of real-time patient interactions while maintaining precise control over performance and privacy.鉴于医疗保健数据的敏感性以及严格的HIPAA合规要求，Hippocratic AI采用了强大的多账户、多集群AWS策略，将生产工作负载与开发环境分隔开来。这种安全且可扩展的基础设施支持数千次实时患者互动，同时对性能和隐私保持精确控制。
Beyond the technical foundation, the real-world impact is profound. Hippocratic AI’s assistants help relieve clinician burnout by handling time-consuming tasks—from surgical prep to post-discharge follow-ups. During a recent hurricane in Florida, the system contacted 100,000 patients in one day, providing medication checks and preventative care—outreach that would be impossible to achieve manually.除了技术基础之外，其现实影响是深远的。希波克拉底人工智能的助手通过处理从手术准备到出院后随访等耗时任务，帮助缓解临床医生的职业倦怠。在佛罗里达州最近的一次飓风期间，该系统在一天内联系了10万名患者，提供用药检查和预防性护理——这样的 outreach 是人工无法实现的。
“With generative AI, patient interactions can be seamless, personalized, and conversational—but in order to have the desired impact, the speed of inference has to be incredibly fast. With the latest advances in LLM inference, speech synthesis, and voice recognition software, NVIDIA’s technology stack is critical to achieving this speed and fluidity.”“借助生成式人工智能，患者互动可以实现无缝化、个性化和对话化——但要产生预期效果，推理速度必须快得惊人。凭借在大语言模型推理、语音合成和语音识别软件方面的最新进展，NVIDIA的技术栈对于实现这种速度和流畅性至关重要。”