Our conviction in Hippocratic AI

At Premji Invest, we are incredibly excited to co-lead Hippocratic AI’s $53M Series A and partner with founders Munjal Shah, Vishal Parikh, Meenesh Bhimani, Alex Miller, Saad Godil, Subho Mukherjee and Debajyoti Datta, alongside the rest of the Hippocratic AI team. We have known Munjal, who is on his fourth entrepreneurial journey, for nearly 2 decades now, and we’ve always been impressed by his knack for identifying new market opportunities early. After spending countless hours with Munjal and the team in their early days when they were whiteboarding different ideas nearly 12 months back, we feel confident about the path they chose. Hippocratic AI’s ultimate vision is to build highly specialized generative AI agents that can engage deeply with patients, providing massive leverage to health systems. When we saw the demo of Hippocratic AI, we were blown away — we could not believe we were talking to a voice-enabled bot.

‍

In the world of AI, we have shied away from applications that are building thin layers on top of large model providers. While we do believe companies will be able to build AI applications with deep moats, the execution of this build will look different than it did in the prior paradigm of software development. We’ve observed a “gap to LLM app” — building specialized vertical applications is exceptionally difficult. Data selection, data curation, model selection, task-specific training/RLHF, and elimination of hallucination are some examples of prominent issues that every entrepreneur needs to solve.

‍

Hippocratic AI falls squarely into our mandate — finding and partnering with exceptional teams that are building n of 1 businesses, solving challenging problems in large markets. We believe the best businesses will utilize highly specialized models that will go 10–15 layers deeper than the best foundation models (OpenAI, Llama, Mistral, Gemini) in providing deeply contextualized answers to questions. Hippocratic AI is doing just this, and they are in the early innings of what we believe is a paradigm shift in software development. We are thrilled for the journey ahead!

Hippocratic AI Overview

Hippocratic AI is building highly specialized voice-enabled generative AI agents for healthcare (e.g. an agent that can conduct a low-risk, non-diagnostic service to a patient about to do a colonoscopy) that communicate with patients and collect patient information (drug dosages, habits, demographic data), conducts follow-ups, executes tasks involving clinical procedures/paperwork, and helps with clinical navigation. In an ideal world, these agents would improve patient compliance and reduce re-admission rates — an acute pain point that clinicians have struggled with for decades. At the end of the day, human nurses have only so much time in their day to follow up with patients to ensure adherence; agents, on the other hand, have “infinite” capacity. The Company’s broader vision is to help hospitals overstaff their nurse capacity, reducing nurse burnout & improving patient satisfaction, both of which result in lower costs and better outcomes for hospitals.

‍

The Company has put a lot of thought into building highly specialized agents. Hippocratic AI had to train their models to ensure their agents communicate concisely (commercial LLMs are often times verbose), demonstrate empathy (commercial LLMs are sometimes rude), navigate social tangents with high EQ (commercial LLMs will remain on tangents), follow checklists (commercial LLMs fail at this), remember important aspects of conversations, and much more. Hippocratic AI’s models are instruction-tuned to ensure the agent can engage in extended conversations, which could involve 40+ “turns”. Initial feedback from human nurses has been overwhelmingly positive — a vast majority of them feel Hippocratic AI’s agents are ready for prime-time: they demonstrate good beside manners, they have high EQ, and they follow hospital/nurse protocols exceptionally well. It is clear that Hippocratic AI is building deep domain-specific IP that will give them a massive edge over competitors.

‍

The team has also been very focused on safety. Without revealing too much about the architecture, the models are HIPAA-compliant, designed to follow hospital protocols (using RAG), over-trained on high-quality data sources, evaluated by clinicians (RLHF conducted by healthcare professionals), and reviewed by a safety governance council, a Physician Advisory Council and a Nurse Advisory Council. Additionally, these agents will avoid diagnoses or recommending drugs/doses that could put patients in danger. How? Hippocratic AI’s models are governed by extremely strict guardrails, and they rely on a built-in “escalation” engine, which loops in live nurses as needed.

Hippocratic AI in Action

James’ much awaited ski trip turned disastrous when a rogue skier caused him to tear his ACL. The diagnosis was grim — limited walking, absolutely no driving and physiotherapy for at least 3 months. A single dad to two middle school girls, James was worried — how was he going to balance getting to work, managing physio appointments, picking up the kids, stocking up the groceries…?

He heard his phone ring.

‍

“Hello?”

‍

“Hi! My name is Rachel. The hospital asked me to check up on you. How are you doing today, James?”

‍

“Oh well, I’m terrified. I don’t understand how I’m going to manage the next 3 months!”

‍

“I understand. ACL recoveries can be very inconvenient. If you’d like, I can help you through it. Don’t worry, James!”

‍

Meet Rachel, Hippocratic AI’s generative AI agent and every patient’s companion and confidante. She can help James not only understand how to take his medication, but also call pharmacies to check and fill-in prescriptions, book and remind him of physiotherapy appointments, deliver his grocery list and arrange play dates for his kids. Hippocratic AI understands that navigating a pre-operation and post-operation life is hard. Rachel is there to help.

Why We’re Excited

We’re excited about Hippocratic AI for a number of reasons. Firstly, given total spend on nursing is ~$400B per year, or roughly 20% of a hospital’s opex budget, the initial TAM is quite large. We believe Hippocratic AI can address anywhere from 20–40% of a nurse’s time spent on administrative work, patient follow-up, and general communication. Assuming for a moment that Hippocratic AI offers a product that is 10x cheaper (on a $/nurse-hour basis) to incentivize adoption, this implies a SAM of $8–16B on the core product alone, without considering any adjacencies.

‍

Secondly, we believe the team has a strong mix of commercial acumen with Munjal Shah (CEO, 3x entrepreneur, 2 exits), Vishal Parikh and Alex Miller, hardware acumen with Saad (CTO, ex-NVIDIA), AI/ML acumen with Subho (Chief Scientist, ex-MSFT & AMZN) and medical experience with Meenesh (ex-COO of El Camino Health, $1.5B health system). Munjal has brought a world-class team together to tackle an exceptionally difficult problem in healthcare.

‍

Finally, we believe Hippocratic AI possesses some powerful competitive advantages that could drive enormous success if everything goes to plan:

‍

Technical IP: there are immense complexities in building a product with low latency and zero hallucination. While we cannot reveal much about the IP, what we can say is that Hippocratic AI has spent a lot of time thinking about domain-specific pre-training and RLHF, experimenting with ensemble models, and figuring out the right composition/weighting of underlying datasets to train their respective models. We believe the future of healthcare chatbots involves building hyperspecialized models purpose-built for conversations in clinical settings.

‍

Capital: Competitors need to be extremely well capitalized to enter this game, given the costs for inference, R&D, and experimentation to “crack” the right model architecture are expensive. Hippocratic AI raised a $67M seed round to attack this problem.

‍

First-mover advantage: given how conservative health systems are, the most difficult part of selling to hospitals is actually getting your foot in the door. But once you’re in, and a health system adopts and trusts Hippocratic AI, it will be extraordinarily difficult for a competitor to unseat them. High switching costs will make this a winner takes most market, similar to what happened with Epic in EHR.

Conclusion

Hippocratic AI’s mission is to help hospitals overstaff their nurse capacity, reducing nurse burnout & improving patient satisfaction, by using AI in an ethical and safe manner. The team is hyper-focused on building the right guardrails to ensure their software is delivered as safely as possible in a clinical setting. We’re early days in understanding the impact of AI in healthcare. Generative AI will supercharge nurses & hospitals to tackle far more problems than they were able to in decades prior. Taking a step back, the types of problems Hippocratic AI is solving extends far beyond nursing alone. Imagine introducing co-pilots for specific types of physician workflows/procedures to reduce burnout for specialist physicians or physician assistants. Nursing is just the beginning — as these underlying models continue to improve with time, we think the sky is the limit. As such, we’re extremely excited about the broader opportunity ahead for Hippocratic AI!