CtrlK
Hippocratic AI is an American artificial intelligence company that develops generative AI agents for healthcare. Founded in 2023 by serial entrepreneur Munjal Shah and a team of physicians, hospital administrators, and AI researchers, the company focuses exclusively on patient-facing, non-diagnostic clinical applications. Its agents handle tasks like post-discharge follow-up calls, chronic disease management check-ins, appointment scheduling, and medication education, but are explicitly prohibited from diagnosing conditions or prescribing treatments.
The company emerged from stealth in May 2023 with a $50 million seed round and has since raised a total of $404 million across three funding rounds, reaching a $3.5 billion valuation by November 2025. Its core technology is the Polaris Safety Constellation Architecture, a multi-model system designed specifically for real-time voice conversations with patients. As of early 2026, the company reported over 180 million patient interactions across more than 1,000 use cases with more than 60 partners worldwide, including Cleveland Clinic, Northwestern Medicine, Ochsner Health, and Universal Health Services.
Munjal Shah studied computer science at the University of California, San Diego, where his senior thesis examined neural networks for predicting protein-ligand binding. He later earned a master's degree in computer science with an AI focus from Stanford University. His first company, Andale Inc., founded in 1999, was one of the earliest cloud-based software-as-a-service platforms, providing e-commerce management tools for sellers on eBay and Amazon Marketplace. Andale was acquired by Vendio in 2004, which later sold it to Alibaba.
Shah's second venture, Like.com, applied AI-based computer vision and machine learning to let consumers search for products visually inside photographs. Google acquired Like.com in 2010 and used the technology as part of its image search infrastructure.
In 2013, Shah co-founded Health IQ, a health-data company that initially offered discounted life insurance to health-conscious individuals by using machine learning to assess fitness-related risk factors. Health IQ later expanded into helping seniors navigate Medicare Advantage plans through what it called a Precision Medicare algorithm. Shah's work at Health IQ gave him direct exposure to the complexities of health insurance, care coordination, and the gaps in patient education that persist throughout the US healthcare system.
By early 2023, Shah concluded that advances in generative AI, particularly the capabilities demonstrated by large language models, had finally made it practical to build AI systems that could hold sophisticated, multi-turn voice conversations with patients. He believed the biggest near-term opportunity was not in clinical decision support for physicians but in the vast number of patient-facing tasks that currently go unmet because there simply are not enough nurses and care coordinators to handle them.
Hippocratic AI was incorporated in early 2023. Shah recruited a founding team that included Meenesh Bhimani, an emergency physician and chief operating officer of El Camino Health in Mountain View, California, who became co-founder and Chief Medical Officer. Other founding executives included Vishal Parikh as Chief Product Officer, Subho Mukherjee as Chief Science Officer, Saad Godil as Chief Technology Officer, Alex Miller as SVP of AI Operations, and Kim Parikh as SVP of Data and Content. The broader founding cohort drew researchers and clinicians from El Camino Health, Johns Hopkins, Washington University in St. Louis, Stanford, University of Pennsylvania, Google, and NVIDIA.
The company's name references the Hippocratic Oath, the ancient Greek medical pledge most associated with the principle of doing no harm. Shah described the name as a deliberate signal that the company's primary commitment was to patient safety rather than capability maximization.
Hippocratic AI launched publicly on May 16, 2023, announcing its $50 million seed round and stating that its early large language model had already outperformed GPT-4 on 105 of 114 healthcare examinations and certifications. The seed round was co-led by General Catalyst and Andreessen Horowitz, with participation from Memorial Hermann, Universal Health Services (UHS), and several other health systems, an unusual feature for a seed round that signaled the investors' intent to validate the technology through direct clinical relationships rather than generic benchmarks.
In December 2023, Hippocratic AI launched an early access partnership program with OhioHealth, Roper St. Francis Healthcare, Evernow, HarmonyCares, and Guidehealth. These partners gave the company real patient interaction data at a stage when most healthcare AI startups were still operating in controlled pilots.
The company also established a Physician Advisory Council in October 2023, bringing in nationally recognized providers to help shape the training and evaluation criteria for its LLM.
| Round | Date | Amount | Lead Investor | Valuation |
|---|---|---|---|---|
| Seed | May 2023 | $50 million | General Catalyst, Andreessen Horowitz | Not disclosed |
| Series A | April 2024 | $53 million | General Catalyst, Andreessen Horowitz | $500 million |
| NVIDIA investment | September 2024 | $17 million | NVentures (NVIDIA), Greycroft, 7Wire Ventures | Not disclosed |
| Series B | January 2025 | $141 million | Kleiner Perkins | $1.64 billion |
| Series C | November 2025 | $126 million | Avenir Growth | $3.5 billion |
Hippocratic AI launched from stealth with a $50 million seed round co-led by General Catalyst and Andreessen Horowitz. The round also included participation from health system investors Memorial Hermann and Universal Health Services, bringing early clinical credibility to the financing. The size of the round was unusual: most seed rounds in healthcare technology fall in the $2-10 million range. Shah explained the larger figure as necessary to build a genuinely safety-validated healthcare LLM, which required substantial compute resources and the cost of using thousands of licensed clinicians for evaluation.
Hippocratic AI raised a $53 million Series A, also led by General Catalyst and Andreessen Horowitz, at a $500 million valuation. The round was accompanied by the announcement of several health system partnerships and the company's first reports of real-world patient interaction data.
In September 2024, NVIDIA's venture arm, NVentures, led a $17 million investment in Hippocratic AI alongside Greycroft and Lee Shapiro of 7Wire Ventures. The deal formalized a technical relationship that had been developing since March 2024, when Hippocratic AI announced a collaboration with NVIDIA to build what it described as super-low-latency empathy inference. NVIDIA made both a financial and engineering commitment: Hippocratic AI would use NVIDIA's inference stack, NVIDIA Riva for automatic speech recognition and text-to-speech, and NVIDIA ACE technologies for voice animation in its patient-facing agents.
On January 9, 2025, Hippocratic AI announced a $141 million Series B led by Kleiner Perkins, with participation from existing investors including Andreessen Horowitz, General Catalyst, Premji Invest, and health system investors NVIDIA, SV Angel, Universal Health Services, and WellSpan Health. The round valued the company at $1.64 billion, making it a unicorn. The company simultaneously announced the launch of its AI Agent App Store for Healthcare. At the time of the Series B, Hippocratic AI reported having completed over one million patient calls.
In November 2025, Hippocratic AI closed a $126 million Series C led by Avenir Growth, bringing its total funding to $404 million and its valuation to $3.5 billion. The round included new investors CapitalG (Google's independent growth fund) and participation from existing investors including General Catalyst, Andreessen Horowitz, Kleiner Perkins, Premji Invest, Universal Health Services, Cincinnati Children's Hospital Medical Center, and WellSpan Health. The company stated it would use the capital to deepen customer deployments, continue developing the Polaris architecture, and pursue strategic mergers and acquisitions.
Polaris is Hippocratic AI's proprietary LLM system, designed from the ground up for real-time, multi-turn voice conversations between AI agents and patients. The architecture differs from single-model approaches by using a constellation of cooperating models that check and constrain each other in real time.
The Polaris system consists of a primary conversational model that manages the flow and emotional tenor of a patient interaction, supported by more than twenty specialist support models. Each specialist model handles a specific subtask: one monitors for potential medication conflicts, another evaluates whether the conversation is escalating toward a clinical emergency, others handle transcription accuracy and multilingual translation. The specialist models run alongside the primary model and can intervene or flag responses before they reach the patient.
Hippocratic AI describes the architecture as a "stateful primary agent" plus specialist agents, a multi-agent approach explicitly designed to reduce hallucination rates and improve medical safety. The paper introducing Polaris, published on arXiv in March 2024 (arXiv:2403.13313), described the system as the first comprehensive clinician evaluation of an LLM specifically optimized for real-time patient voice conversations.
The original Polaris system was approximately one trillion parameters in total across all its models. Polaris 3.0, released in March 2025, grew to 4.2 trillion parameters across 22 LLMs.
Hippocratic AI trained Polaris on a combination of proprietary datasets, clinical care plans, healthcare regulatory documents, and medical reasoning materials. A distinctive element of the training process involved extensive use of patient actor simulations and conversations with experienced nurses to develop what the company calls rapport building, trust building, empathy, and bedside manner as explicit training objectives rather than incidental byproducts. The company also incorporated feedback from real patient interactions gathered through its early access program.
Before any version of Polaris goes into production, Hippocratic AI runs it through evaluation sessions conducted by US-licensed clinicians who role-play as patients. Evaluators assess medical safety, clinical accuracy, conversational quality, and the agent's ability to recognize when a patient needs escalation to a human clinician.
| Version | Release | Clinical Accuracy | Parameters | Patient Satisfaction |
|---|---|---|---|---|
| Polaris 1.0 | 2024 | 96.79% | ~1 trillion | 8.56/10 |
| Polaris 2.0 | Late 2024 | 98.75% | Not disclosed | 8.72/10 |
| Polaris 3.0 | March 2025 | 99.38% | 4.2 trillion (22 LLMs) | 8.95/10 |
| Polaris 5.0 | April 2026 | 99.90% correct clinical advice | Not disclosed | Not disclosed |
Polaris 3.0 introduced Deep Thinking Models that triple-check labs, medications, and escalation triggers using an offline reasoning pass before delivering a response to the patient. The release also added multilingual support covering nine non-English languages, with Spanish achieving a 99.83% accuracy rate.
Polaris 5.0, announced in April 2026, was built on a dataset of over 180 million patient interactions and was reported to outperform major frontier models on a range of medical tasks while maintaining a 99.90% correct clinical advice rate with zero recorded severe harm events across its deployment history.
In November 2024, the US Patent and Trademark Office issued a patent for Hippocratic AI's agentic AI clinical safety innovation, covering aspects of the Polaris constellation approach to clinical error reduction.
Hippocratic AI's most fundamental design constraint is the prohibition on diagnosis and prescription. The company explicitly built its agents so they cannot and will not tell a patient what condition they have, recommend a specific drug, or comment on clinical decisions made by their physician. When a patient asks a diagnostic question, the agent is trained to redirect to their care team.
Additional restrictions include: the agents do not provide hospice care guidance, do not address mental health disorders (beyond connecting patients to appropriate resources), and do not handle patients under two years old. These limits are not configurations that customers can override; they are baked into the Polaris training and architecture.
Shah has repeatedly described this scope limitation as a deliberate strategy: by staying out of the highest-risk clinical decisions, Hippocratic AI can deploy agents at scale without requiring the same regulatory approval pathway as a clinical decision support tool. The Food and Drug Administration (FDA) classifies clinical decision support software on a risk spectrum, and software that directly influences a clinician's diagnostic or treatment decision requires more extensive regulatory clearance than software that provides patient education or care coordination.
In early 2025, Hippocratic AI published the Real-World Evaluation of Large Language Models in Healthcare (RWE-LLM) framework, simultaneously releasing a preprint on medRxiv. The framework draws from red-teaming methodologies used in AI safety research and adapts them to the specific demands of clinical deployment.
The RWE-LLM approach operates across four phases: pre-implementation testing, tiered clinical review, error resolution, and continuous monitoring. The evaluation process employs a three-tier review structure in which errors are categorized by severity and routed through escalating levels of clinical oversight.
For the validation of Polaris 3.0, Hippocratic AI engaged 6,234 US-licensed clinicians (5,969 nurses and 265 physicians) averaging 11.5 years of clinical experience. These clinicians evaluated over 307,000 unique patient calls. The company also maintains a standing network of more than 7,500 licensed clinicians for ongoing validation work.
In July and August 2025, Hippocratic AI achieved HITRUST e1 certification for its full agentic platform, a healthcare information security certification that demonstrates compliance with HIPAA and other regulatory requirements relevant to patient data handling.
On January 9, 2025, concurrent with its Series B announcement, Hippocratic AI launched the AI Agent App Store for Healthcare. The store is a marketplace where US-licensed clinicians can design, publish, and monetize AI agent workflows built on the Polaris infrastructure.
Clinicians can build a new agent workflow in as little as 30 minutes without writing any code. After internal testing by the creator, Hippocratic AI staff review the agent's clinical safety. The company also convenes a panel of 6,000-plus nurses and 300-plus physicians to evaluate agents before they are approved for customer deployment. Only agents that pass this clinical review are listed in the store.
Once listed, when a Hippocratic AI customer deploys the agent, the clinician creator receives 5% of the base usage fees plus 70% of any premium rate the creator sets above the base fee. This model is designed to align the interests of clinical creators with the quality and accuracy of their agents, since creators share financially in deployment outcomes.
The App Store launched with more than 300 AI agents covering 25 medical specialties. Representative agent types include cervical cancer check-ins, postpartum mental health monitoring, wound care follow-up, diabetes screening, post-surgical recovery calls, vaccination outreach, and pre-authorization assistance.
By the time of the Series C in November 2025, the company reported over 1,000 clinical use cases available through the platform.
Hippocratic AI's core product is a voice-first AI agent that conducts outbound and inbound patient calls on behalf of health systems, payers, and pharmaceutical companies. The agents operate under the patient's care team name (for example, WellSpan Health named its agent Anna) and are designed to sound empathetic, consistent, and clinically accurate. Calls can be asynchronous outreach (the agent calls a patient 48 hours after discharge) or synchronous responses (a patient calls in and is connected to an AI agent for a routine inquiry).
Use cases fall into three broad customer segments:
Health systems and providers: Post-discharge follow-up calls to reduce readmission rates, chronic disease management check-ins (diabetes, hypertension, heart failure), surgical preparation and recovery education, vaccination campaign outreach, appointment scheduling and reminders, patient intake history collection, and social determinants of health screening.
Payers (insurance companies): Eligibility and benefits confirmations, medication reconciliation calls, case management outreach, annual health risk assessments, pharmacy management support, and member experience surveys.
Pharmaceutical companies: Patient education about support programs, pre-screening for clinical trial enrollment, real-world evidence collection through longitudinal patient follow-up, and above-brand disease education.
Announced in April 2026, the AI Front Door is an inbound patient access product. It handles incoming calls from patients seeking appointments, referrals, or routine information, routing them to the right clinical resource or completing the interaction autonomously when no human escalation is needed. Cleveland Clinic and OhioHealth helped shape the product during its development.
Also announced in April 2026, the Nurse Co-Pilot is an AI voice assistant built for inpatient nursing workflows. Unlike the outbound patient-facing agents, the Nurse Co-Pilot is designed to run at the bedside, handling tasks that currently require nurse time but do not require clinical judgment: admission education (walking a newly admitted patient through what to expect), ongoing patient education during a stay, discharge preparation, and caregiver engagement. The product was co-developed with Cincinnati Children's Hospital Medical Center, OhioHealth, and Cleveland Clinic, who contributed nursing leadership to the design process. The company claimed the tool could return one to four hours of nursing time per shift.
Hippocratic AI's agents operate primarily through voice, and the latency and naturalness of voice interactions are central to the product's clinical effectiveness. A patient who experiences significant pauses, robotic prosody, or transcription errors during a post-discharge call is less likely to provide accurate answers or remain engaged.
In March 2024, Hippocratic AI announced a collaboration with NVIDIA to develop what it described as "super-low-latency empathy inference." The partnership involved integrating NVIDIA Riva for automatic speech recognition and text-to-speech conversion, and NVIDIA's broader AI infrastructure for low-latency inference serving. The collaboration was designed to bring AI agent voice interaction latency below the threshold at which humans typically notice a delay in conversation.
NVIDIA also contributed NVIDIA ACE (Avatar Cloud Engine) technologies for realistic voice synthesis and animation in cases where agents are presented with a visual avatar interface. The partnership was later formalized with NVIDIA's venture arm NVentures taking a $17 million stake in the company in September 2024.
The NVIDIA collaboration gave Hippocratic AI access to inference optimization expertise and hardware that would have been difficult to develop independently, and it gave NVIDIA a prominent design win in healthcare AI, a sector where NVIDIA had been seeking to demonstrate the clinical relevance of its GPU infrastructure.
By November 2025, Hippocratic AI reported partnerships with more than 50 health systems, payers, and pharmaceutical companies across six countries. Notable US health system customers include Cleveland Clinic, Northwestern Medicine, Ochsner Health, Moffitt Cancer Center, University Hospitals, Cincinnati Children's Hospital Medical Center, Sanford Health, OhioHealth, Memorial Hermann, WellSpan Health, HonorHealth, and Universal Health Services. International partners include Guy's and St Thomas' NHS Trust in the United Kingdom, Sheba Medical Center in Israel, Burjeel Holdings in the UAE, Cleveland Clinic Abu Dhabi, and Fraser Health in Canada. In May 2025, Hippocratic AI partnered with EUCALIA Inc. to deploy the first Japanese-language patient-facing AI healthcare agent.
The company has also signed consulting and deployment partnerships with KPMG (announced July 2025), BCG (January 2026), Huron Consulting Group (January 2026), and CTG (November 2025), organizations that help health systems assess and implement the platform.
WellSpan Health, one of the earlier production deployments, publicly described satisfaction with the product's safety profile. The health system's chief executive cited the accuracy of its safety information and the agent's ability to stay within appropriate clinical scope.
Reported aggregate outcomes across deployments include: a 12-times average return on investment, a 30% reduction in hospital readmissions for engaged patients, and a 60% completion rate for vaccination outreach goals.
Hippocratic AI occupies a specific niche in the healthcare AI market. It is neither a clinical decision support tool for physicians nor a general-purpose enterprise AI agent. Its closest competitors are other patient-facing engagement platforms, though several of those were built on general-purpose LLMs rather than healthcare-specific architectures.
| Company | Focus | LLM Approach | Diagnostic Capability | Voice-First | Clinician Validation |
|---|---|---|---|---|---|
| Hippocratic AI | Patient-facing non-diagnostic agents | Proprietary healthcare-specific (Polaris) | Explicitly prohibited | Yes | 7,500+ licensed clinicians |
| Sierra AI | General enterprise customer-facing agents | Proprietary, not healthcare-specific | Not applicable | Yes | Not healthcare-specific |
| Microsoft Nuance DAX Copilot | Physician documentation and ambient AI | GPT-4 based, integrated with Epic | Indirect (documentation support) | Yes (ambient) | Via Epic integration |
| OpenAI (via API) | General-purpose; customers build own healthcare applications | GPT-4/o3 | Possible but uncontrolled | Via third-party integration | No purpose-built validation |
| Google Health AI | Clinical imaging, diagnostic support | Med-PaLM and Gemini variants | Yes (imaging, triage) | Limited | Research-focused |
Sierra AI, founded by former Salesforce CEO Bret Taylor, offers voice and text AI agents for enterprise customer interactions. Its platform competes with Hippocratic AI in the voice agent infrastructure layer, but Sierra is healthcare-agnostic and does not impose the non-diagnostic constraints or clinician validation process that Hippocratic AI's architecture enforces. Sierra is a more general purpose platform; Hippocratic AI's value proposition rests on the safety architecture and clinician validation being built in rather than left to the deploying customer.
Microsoft's Nuance DAX Copilot focuses primarily on reducing physician documentation burden through ambient AI that listens to clinical encounters and generates structured notes. This is largely a different workflow than Hippocratic AI's outbound patient engagement, though both companies are expanding, and the categories are beginning to overlap as Hippocratic AI enters inpatient nurse workflow tools.
Patients who leave a hospital after a procedure or acute illness are at elevated risk of readmission within 30 days, particularly if they do not understand their discharge instructions, fail to pick up medications, or do not recognize warning signs. Hippocratic AI agents conduct structured follow-up calls in the 24-72 hours after discharge, reviewing medication adherence, symptom monitoring, and appointment scheduling. The company reported that health systems using this use case saw a 30% reduction in readmission rates among patients who received AI agent follow-up calls.
Patients with diabetes, hypertension, heart failure, or chronic obstructive pulmonary disease require regular check-ins to monitor symptom changes, review medication adherence, and reinforce lifestyle recommendations. These check-ins are clinically valuable but time-intensive for nursing staff. Hippocratic AI agents conduct these calls on behalf of care management programs, escalating to a human clinician when a patient reports symptoms that exceed pre-defined thresholds.
Health systems and payers run annual vaccination campaigns for influenza, COVID-19, RSV, and other preventable illnesses. These campaigns require large numbers of outbound calls to patients who are due or overdue for vaccination. Hippocratic AI agents can conduct these calls at scale, answer common questions about vaccine safety and availability, and schedule appointments. The company reported a 60% vaccination goal completion rate across campaigns using its agents.
Insurance companies use Hippocratic AI agents for member outreach around annual health risk assessments, a regulatory requirement under Medicare Advantage contracts, as well as for eligibility confirmations and medication reconciliation calls. These interactions are administratively important but low-acuity, making them well-suited for AI agent handling.
Pharmaceutical companies use Hippocratic AI agents to conduct initial pre-screening interviews with patients who may be eligible for clinical trials. The agents can administer standardized screening questionnaires, collect patient history relevant to trial eligibility criteria, and schedule patients for a follow-up call with a human coordinator if they pass the initial screen. This use case reduces the manual burden of trial enrollment while maintaining human oversight of the actual enrollment decision.
Hippocratic AI received broadly positive coverage from healthcare technology media. The January 2025 Series B announcement was covered by TechCrunch, Fierce Healthcare, MobiHealthNews, and The Healthcare Technology Report, with most coverage focusing on the unicorn valuation and the simultaneous App Store launch as signals of market validation.
Health system partners have been publicly enthusiastic. WellSpan Health's CEO cited the safety profile specifically. OhioHealth's leadership described the company's safety validation process as aligned with their own clinical standards. Cleveland Clinic's participation in co-designing the Nurse Co-Pilot indicated a level of institutional engagement beyond a typical vendor relationship.
CB Insights named Hippocratic AI to its Digital Health 50 list of most promising startups in December 2024. Forbes included the company in its America's Best Startup Employers 2026 list.
KPMG and BCG's decisions to build formal consulting partnerships around deploying Hippocratic AI's platform were interpreted by industry observers as signs that the company had crossed from early-adopter to mainstream enterprise consideration. Consulting partnerships of that type typically follow a period of demonstrated customer success at scale.
Hippocratic AI's non-diagnostic mandate, while central to its safety strategy, also limits the range of clinical tasks its agents can address. Patients frequently move from asking about their medications to asking what a symptom means. When an agent reaches the boundary of its permitted scope, it must redirect, which can feel abrupt to patients used to more comprehensive answers from human nurses or internet searches. The company acknowledges this trade-off and has framed it as a feature rather than a bug, arguing that an agent that clearly defers on diagnostic questions is safer than one that attempts an answer.
The primary interaction model is voice, which creates access constraints for patients who are deaf or hard of hearing, who do not speak one of the supported languages, or who are in environments where a phone call is not practical. As of the Polaris 3.0 release, the system supports nine non-English languages, but the majority of available agents were designed for English-speaking patients.
Hippocratic AI's clinician validation network, while large, focuses on nurses and physicians in the United States. Healthcare norms, clinical terminology, and patient communication expectations differ across countries, and agents deployed internationally (in Israel, the UAE, Canada, and Japan, as of 2025) required additional localization and validation. The company has not published peer-reviewed outcomes data from international deployments.
Wider concerns about AI in patient-facing healthcare apply to Hippocratic AI's platform as well, even if the company has not been the subject of specific public criticism. These include: the possibility that patients may over-rely on AI agents and delay contacting a human clinician when they should; privacy and data residency issues when patient conversations are processed by third-party AI systems; and the challenge of keeping AI agents current when clinical guidelines change. Hippocratic AI addresses the last concern through its continuous monitoring framework, but the company does not publish detailed information about how quickly guideline updates propagate through its deployed agent population.