Click HERE to request case studies

  • Home
  • Meet Flo
    • Who we serve

      Our integrated care pathways deliver lasting change, meaningful outcomes, and cost savings to our partners.

    • Pathways & Programmes

      Proven to deliver better clinical outcomes and free up time and resources for your healthcare team

    • Evidence & Insight

      Clinical evidence and case studies show the value Florence provides to our partners.

    • About Us

      Florence delivers precise, psychology-based, intelligent health messaging.

  • Contact
  • Request Demo
    • English US
Close this search box.



Groundbreaking AI Research Partnership: Synthetic Patients and Gen AI Empowered ‘Nurse’ 

AI Twins

Generated Health and William & Mary have announced a groundbreaking agreement in the research and development of next generation AI to further ‘Automation of Care Coordination.’  The team will create ‘synthetic patient’ data to train the digital nurse ‘Florence’ using Generative AI and Reinforcement Learning.

Florence automates the engagement and management of patients through clinical conversations via text.  “Our mission is to create 1 million digital nurses that enable better, more accessible healthcare at lower cost,” said Ingolv Urnes, CEO of Generated Health, Inc.  “We already have extensive clinical evidence across 200,000 patients demonstrating strong health outcomes while freeing up clinical time.  We have proven we can combine clinical protocols and AI to great effect.  We are now developing the next generation.”  

“We will be leveraging the unique data set that Generated Health has and effectively developing two models; a diffusion model that creates ‘synthetic patient’ data, which in turn will be used to train the nurse model combining Large Language Model (LLM) and Reinforcement Learning,“ explained Dr Haipeng Chen, assistant professor of data sciences at William & Mary and a previous CRCS postdoctoral fellow of Computer Science at Harvard University.

“My primary interest is AI for social good,” said Dr Chen. “Given the challenge of access to care and crippling staffing shortages, this project should make a meaningful contribution to millions of patients in the US suffering from chronic conditions.”

Goal: Autonomous AI Nurse, Fully Personalized, Dynamic Decision Making

“The goal is to develop a digital nurse that is intelligent and capable of effective decision making with a minimal set of absolute clinical rules, that uses Reinforcement Learning and truly treats each patient as an individual,” according to Mr Urnes.

“It is universally recognized that the ability to engage, educate, and provide frequent feedback to patients is critical to ensure that care plans, including medication regimens and life-style changes, are followed.  Obviously, the nurse – human or digital – must deliver ongoing care that is as personalized as possible and at all times based on the patient’s disease progression”.

By developing this next generation, Florence’s ability to drive patient engagement will be further enhanced, including ‘her’ ability to leverage sophisticated behavior change techniques such as motivational interviewing.

Existing Clinical Evidence and Unique Training Data   

This research project will leverage Generated Health’s unique data set covering over 25 million clinical conversations via text with over 200,000 patients in the US, UK, and Australia.  Generated Health has extensive, validated clinical evidence across all key chronic conditions, including diabetes, hypertension, COPD, CHF and multimorbid patients. 

For example, in the management of hypertension and medication titration, Florence was able to sustainably reduce average systolic blood pressure by 15%, reduce physician and pharmacist time by over 75%, and reduce administrative work ten-fold.  Florence has also proven valuable in managing patients (typically with chronic conditions) pre- and post-procedure or surgery; in cardiac surgery an independent study concluded that Florence reduced hospital readmissions by 67% compared to the control group.  

Synthetic Patients: Better, High-Volume AI Training Data

Currently Florence AI uses a public LLM to interpret ‘patient intent’ and to build a patient profile. De-identified incoming messages from patients are processed iteratively combining proprietary rules and prompt engineering.  Florence’s outgoing messages and escalation triggers are locked down in clinical protocols to eliminate the risk of hallucination.  Florence AI testing is based on historical data and a human evaluation team.

To speed up the development of a more Autonomous AI Nurse, more training data and high-volume simulations are required. Therefore, the teams from William & Mary and Generated Health are now developing an AI diffusion model to simulate patient engagement with Florence.   This generative model is trained on de-identified real-world data and will be able to create synthetic patient data with statistical properties similar to those of real patients.

By introducing conditions into the diffusion model the team will also be able to simulate various scenarios such as patients with steadily increasing blood pressure or rapid changes in mood score to optimize Florence’s next-best-actions using Reinforcement Learning (see below).

The key advantage to synthetic patient data is safety and the ability to simulate real-world behavior.  Another benefit is the absence of privacy issues which will enable the team to share the synthetic data with other researchers.

Combining Generative AI and RL Fully Personalized and Dynamic Patient Intervention   

Leveraging synthetic patient data, Florence will be trained through high-volume interaction with a population of synthetic patients using an AI approach called Reinforcement Learning (RL).

RL is a type of machine learning paradigm where an agent (i.e., the digital nurse) learns to make decisions by interacting with an environment (i.e. patients and clinical workflows).  In RL, the agent receives feedback in the form of positive or negative rewards based on the actions it takes with the objective of learning strategies that maximize the cumulative reward over time. This approach has been successfully adopted in training popular AI models like ChatGPT and AlphaGo.

Eventually, the trained RL agent (in this case, Florence) will then engage with real patients, and continuously self-improve via feedback from the interactions with real patients and clinicians.

About Generated Health

Generated Health’s software platform Florence automates the engagement and management of patients through clinical conversations via ‘simple’ SMS text.  The dialogs with patients are powered by evidence-based protocols and AI.  The Company was spun out of an R&D program within the UK’s National Health Service in 2021 and relocated to the US in 2023.  

To date, Florence has managed over 200,000 patients across the US, UK and Australia conducting over 25 million clinical conversations.  Over 500 clinicians have contributed to the creation of clinical protocols.  Florence is proven to deliver better patient experience, improve clinical outcomes, and save significant clinical time. 

Examples include: 

  • Hypertension: Clinical time down by 75%, ten-fold reduction in admin tasks, 97% medication adherence, sustained BP control 
  • Diabetes: 64% less time spent by clinicians, A1c reduced by 1 point 
  • Heart failure: Hospital admissions reduced by over 65% 
  • Patient Net Promoter Scores consistently above 75% 

For further information, please contact: or visit or follow @GeneratedHealth on LinkedIn

About William & Mary and Dr Haipeng Chen

The College of William & Mary is renowned for blending traditional academic values with cutting edge research, positioning it as a distinguished institution on the global stage. Its commitment to fostering a dynamic environment where research thrives is evident across a wide range of disciplines.

Haipeng Chen is assistant professor of data science at William & Mary.  Previously, he was a postdoctoral fellow at Harvard University and Dartmouth College. He obtained his Ph.D. from Nanyang Technological University (NTU). His primary research interest is in AI for social good. He investigates theoretical and practical research questions in deploying AI techniques (e.g., reinforcement learning, combinatorial optimization, and prediction) to social domains (e.g., health, environment). His research has been recognized as the best paper nomination at AAMAS-2021, Innovation Demonstration Award runner-up at IJCAI-2019, and Champion of the 2017 Microsoft Malmo Collaborative AI Challenge. He publishes in premier AI/data science conferences such as AAAI, IJCAI, ICLR, NeurIPS, AAMAS, UAI, KDD, ICDM, and journals (e.g., IEEE/ACM Transactions). He also regularly serves as reviewer/meta-reviewer at these AI conferences. He co-chaired several workshops on the theme of AI and society, including two at ICLR and IJCAI, and one at Harvard. Aiming at algorithm deployment, he collaborates with non-profits and practitioners from the field, including the Virginia Department of Health, Tufts Medical Center, Lackey Clinic, The Family Van, Mobile Health Map, Safe Place for Youth, and Wadhwani AI.

For further information, please contact:

Dr Haipeng Chen       Webpage:

Florence Logo

Case Study Packs

Contact Us

Generated Health
Mercury House, 117 Waterloo Road
London SE1 8UL

Call: +443455050120