NVIDIA AI Podcast Ep. 262

Hippocratic AI’s Munjal Shah on How AI Agents Are Expanding Healthcare Capacity

Date: June 25, 2025
Host: Noah Kravitz (at NVIDIA GTC 2025)
Guest: Munjal Shah, Co-founder & CEO, Hippocratic AI

Episode Overview

In this episode, Noah Kravitz sits down with Munjal Shah, CEO of Hippocratic AI, a startup building large language model (LLM)-powered AI agents for healthcare with a strong focus on safety. The conversation explores Hippocratic AI’s novel approach to augmenting healthcare capacity through scalable virtual AI clinicians, the launch of their Healthcare AI Agent App Store, their unique technical architecture, and how AI can usher in a new era of "Healthcare Abundance." Shah provides candid insights into infrastructure challenges, clinician involvement, safety protocols, and what’s next for both the company and AI in healthcare.

Key Themes & Discussion Points

Introducing Hippocratic AI and Its Mission

• [00:54] Shah introduces Hippocratic AI as a “safety-focused large language model focused on healthcare,” specifically designed to create AI clinicians—agents that reach out to patients, such as by calling post-surgery to check on their recovery, medication needs, and more.

  “It’s really an agent that talks to patients and delivers care.”
  — Munjal Shah ([00:54])

• Founded: ~Two years ago ([01:21])

• Production scale: 1.85 million patient calls completed as of the episode ([01:27]).

Patient Experience & Response

• [01:40] On reactions to receiving healthcare from an AI agent:

  • Average patient rating: 8.95 out of 10
  • Initial skepticism: 30% hesitant to talk to AI, but with gentle rebuttal from the AI, 85% end up engaging.

  “Within like 30, 60 seconds, when they realize this is not your grandfather’s IVR ... and is empathetic, they just talk away.”
  — Munjal Shah ([01:51])

  “This thing listens to every word and responds ... and that’s gold in the modern age.”
  — Munjal Shah ([02:25])

The Vision: “Age of Healthcare Abundance”

• [02:54] Shah describes the shift from triage and scarcity to a world where AI agents enable “clinical abundance,” offering scalable, routine care and monitoring.

  “All of healthcare is premised on this idea of clinical scarcity ... Instead of, how do we get to a place where it’s infinitely abundant?”
  — Munjal Shah ([02:55])

• AI agents enable novel interventions, such as rapidly assessing vulnerable patients during a heatwave, which would be impossible to scale with human staff alone ([04:23]).

AI and Human Clinicians: Complementary Roles

• [04:39] The current technology excels at virtual care; human clinicians continue to provide necessary physical and higher-complexity medical intervention.
  • AI frees up clinicians to focus on complex, in-person care by automating repetitive, scalable patient touchpoints.

Hippocratic AI’s App Store for Healthcare Agents

• [05:11] Hippocratic AI launched an AI Agent App Store, leveraging their LLM foundation (“making a new use case takes four minutes” [05:31]) and inviting clinicians nationwide to author new care scripts.

  “If you worked in a concussion clinic as a nurse for the last 20 years ... why don’t you write a new script and put your script in our app store? ... Once it’s live, you’ll get paid a portion of all the revenue it makes.”
  — Munjal Shah ([06:18])

• Only licensed U.S. clinicians can contribute; every new use case is safety-tested and validated before deployment.

Technical Deep Dive: Inference and System Architecture

• [07:00] Shah highlights the importance of inference (as opposed to only training):

  • Their system uses 22 models in conjunction—one primary 400B LLM, 19 safety supervisors, and 2 “deep thinking” models for double-checking.
  • Each patient interaction runs against all these models to maximize safety and clinical appropriateness.

  “That’s a lot of inference. That’s 22 models of inference. It’s 4.2 trillion parameters we’re running every time.”
  — Munjal Shah ([07:53])

• Infrastructure scale: Their current instance uses over 128 Nvidia H100 GPUs just to load into RAM ([08:02]).

Latency and Infrastructure Challenges

• [09:25] Real-time voice communication imposes a strict latency requirement (1.5 to 2 seconds end-to-end), demanding techniques different from text-based LLM interfaces.

  “Our inference isn’t a matter of trying to optimize cost per token, but trying to optimize latency.”
  — Munjal Shah ([10:07])

• Partnership with NVIDIA and tuning of open-source inference engines for latency, not just throughput ([10:50]).

The “Constellation Architecture”

• [11:05] Their models operate as a constellation, each specialized (e.g., an “overdose engine”) actively monitoring conversations for specific risks or medical needs.

  “We literally have multiple models double-checking each other ... RLM knows how to ask all these questions and knows how to navigate assessing whether it’s actually an overdose.”
  — Munjal Shah ([13:10])

Safety and Clinical Validation

• [13:34] Instead of just dataset transparency, Hippocratic AI conducts rigorous “output testing:”

  • 6,000 licensed clinicians participated in 309,000 clinical test calls, acting as patients and marking every error.
  • Iterative improvement based on real-world feedback, resulting in unprecedented levels of validation.

  “We have 6,000 US licensed clinicians who have now done 309,000 clinical test calls ... We call this output testing. We’ve done more output testing than anybody.”
  — Munjal Shah ([14:45])

  “If this is going to call my mother, she’s 81 years old, I want to know what it’s going to say ... I’m not comforted by, ‘Oh, I trained it on this.’”
  — Munjal Shah ([15:12])

Clinician and Health System Reaction

• [16:45] Most clinicians see AI augmentation as necessary due to acute staff shortages—especially post-pandemic.

  • Rapid health system adoption: 25 health system/provider/pharma clients signed in 6-7 months; expecting 30–40 by next June, a rate unheard of in healthtech ([17:54]).

  “The needs [are] recognized ... there’s a lot of pain around staffing and staffing shortages.”
  — Munjal Shah ([18:18])

Roadmap: Payers, Pharma, and Global Expansion

• [18:28] Plans to:

  • Expand into payer (insurance) support: automating case management touchpoints
  • Serve pharma: clinical trial compliance, qualification, and recruitment
  • Expand internationally: launched in UAE, soon in Southeast Asia

  “The whole world is short. And you know, with the aging population ... we all have no choice.”
  — Munjal Shah ([19:42])

Notable Quotes & Memorable Moments

• “[AI agents] speak every language, they remember every conversation. They’re clinically safe and they can take care of everybody at all times.”
  — Munjal Shah ([03:08])

• “We're crowdsourcing, but only from clinicians. You actually have to send us your license number.”
  — Munjal Shah ([06:38])

• “Latency is everything. ... In a voice conversation you have a 1.5 to 2 second budget end-to-end.”
  — Munjal Shah ([09:47])

• “Who knows best how to assess an AI, except the clinicians.”
  — Munjal Shah ([15:22])

• “We’re crossing a wood plank bridge and the planks are showing up like two seconds before we hit the next step.”
  — Munjal Shah, on technical disruption ([15:35])

Key Timestamps

| Timestamp | Segment/Topic | |------------|-------------------------------------------------| | 00:54 | What is Hippocratic AI and AI clinicians? | | 01:40 | Patient reactions to AI-delivered healthcare | | 02:54 | “Age of Healthcare Abundance” explained | | 04:39 | AI’s role vs. human clinicians | | 05:11 | Launch of Healthcare AI Agent App Store | | 07:00 | Move from training to inference in AI | | 08:02 | Infrastructure and scale (Nvidia H100 GPUs) | | 09:25 | Latency and real-time voice challenge | | 11:05 | Constellation architecture and redundancy | | 13:34 | Clinical testing and model validation | | 16:45 | Clinician acceptance and health system adoption | | 18:28 | Roadmap: payers, pharma, international expansion | | 20:06 | Research & safety testing publications |

Further Information

• Learn More: hippocraticai.com
  • Details on their LLM, architecture, published safety protocols, and company values.

Takeaways

• Hippocratic AI is at the forefront of using LLM-based virtual agents to safely and scalably augment healthcare, with a deeply clinician-oriented approach to safety, validation, and product development.
• Their focus on “healthcare abundance” reframes AI as a tool for positive systemic change, addressing global clinician shortages rather than replacing providers.
• Technical innovation in inference, latency, and modular safety architectures are key to their real-time AI agent deployments.
• Open collaboration and clinician-driven customization (via the App Store) enable broad, context-aware healthcare support at scale.
• The industry is responding rapidly to Hippocratic’s vision, signaling a major shift in how AI will expand access and improve care globally.

Episode Host: Noah Kravitz
Guest: Munjal Shah, Hippocratic AI
For a deeper dive, downloadable papers, and more, visit hippocraticai.com.