What is Felix?

Felix is a personalized, digital care manager that helps patients receive concierge-level care from their physician's office at a low cost. We connect with patients using an SMS interface or Amazon Echo app by asking them about general wellness, tracking patient responses to increasingly personalize interactions, and using tone of speech analysis as well as support for contextual questions to naturalize conversations. This includes monitoring their blood sugar and giving real-time feedback on reported values, sending medication reminders, and providing a comprehensive 16-week diet and exercise program (modeled off DPP). A risk assessment algorithm tracks patient-reported symptoms and finds trends in health data to notify PCPs when a patient may be at risk of immediate hospital readmission or is generally performing worse, prompting their attention.


Felix is built using NLP techniques that are yet to be seen in healthcare from both a development and deployment standpoint. We use both SMS and voice-activated tech like Echo/Google Home to rapidly scale delivery to the highest risk, elderly population that may not be comfortable with app-based solutions and can therefore interact with Felix as if it's a virtual nurse at their doctor's office. Additionally, we can reach individuals in lower socioeconomic brackets by using SMS to reach those who may not have smartphones or familiarity with other tech. Using a core AI engine to drive all of our conversations with patients, there is tremendous sustainability in expanding both the breadth and depth of the content we support. Additionally, content delivery to patients is sustainable across large populations given that we use SMS, which is highly accessible, and have infrastructure to support thousands to millions of conversations in parallel.


Felix is not a traditional conversational agent such as the core Alexa agent or Siri in the sense that Felix can maintain a long-term conversation. This includes contextual conversations, data-driven responses (i.e. sending diet guidance immediately after a patient reports an elevated blood sugar), and tone of speech analysis. We use a series of Watson classifiers but have developed several pre-processing steps to do speech analysis and sentence deconstruction. We've collected nearly 100,000 data points thus far and used them to train our AI, which has now achieved approximately 80% accuracy for T2DM, far better than similar interventions in other industries. Additionally, we've used existing literature, CDC data, and our own training data to develop a risk assessment algorithm to assess a patient's readmission probability, something that health systems currently don't have. This has helped prevent nearly 150 readmissions thus far.


A 2016 meta-analysis of academic studies pooling over 2700 patients showed that SMS reminders nearly doubled both short- and long-term medication adherence and patient motivation in T2DM patients across the socioeconomic spectrum. SMS-based health interventions such as ‘Text2Move’, which reminded patients with diabetes to maintain an active lifestyle over six months, resulted in a 1% reduction in hemoglobin A1c.

Felix is substantially more comprehensive and accessible to patients; on average, a cohort of 1,000 patients who have used Felix thus far engaged with it 87% of the time for atleast 30 days. This is largely due to the fact that we’ve developed Felix to converse with patients instead of just sending them reminders, a testament to the quality of the tech and usage of SMS/Echo. Felix has sent nearly 90,000 messages to-date, prevented 150 ED visits related to blood sugar control, and drastically improved glycemic control using a combination of medication reminders and a 16-week coaching program.


We've developed Felix over the past two years with the help of endocrinologists at Brigham & Women's Hospital and worked to collect data to make it self-sustained. We've proudly gone from a response accuracy rate of 30% 18 months ago to >80% today and have effectively implemented this for nearly 1,000 patients across the country from Menlo Park to Phoenix to Atlanta to Boston. Our risk algorithm has identified and prompted preemptive attention to prevent 148 readmissions to-date and patients are engaging with Felix at a rate of 87% over the long-term.


Nisarg Patel - https://www.linkedin.com/in/nisargp/ Manav Sevak - https://www.linkedin.com/in/manavsevak/ Kunaal Naik - https://www.linkedin.com/in/kunaalnaik/

Nisarg is a fourth-year student at Harvard Medical School, Manav is a fourth-year undergraduate at Georgia Tech studying biochemistry and computer science, and Kunaal is a fourth-year undergraduate at Georgia Tech studying computer science. Additionally, we've brought on Manjinder Kandola who is a first-year Stanford MBA student, Kurt Carpenter who is a software engineer at Microsoft, Sam Marder who is a software engineer at Google, and Malhar Patel who is a third-year EECS student at Berkeley. Between our core team and advisory board, we have a highly interdisciplinary team with knowledge of both the clinical and technical aspects of implementing this kind of intervention and effectively scaling it.


Current funding includes $135,000 in grants from the NIH, Harvard Business School, Georgia Tech, and Partners HealthCare. We currently have three active, paying clients who contribute to our monthly revenue as well.

5-Year Plan

We currently have a MRR of approximately $3,000 with nearly five secured implementations of Felix for diabetics being serviced through private practices. We anticipate scaling this to a market of nearly 100 practices within the next 12 months and want to encompass the entire market of nearly 31,000 diabetes-focused practices operating at a loss within 5 years. In this timeframe, we anticipate expanding the scope of our application to include several co-morbidities, type 1, and other conditions to maximize our reach to all 30 million diabetics in the United States. We envision Felix to become a comprehensive clinical assistant that can entirely oversee post-discharge care for a diabetic patient and therefore essentially become an added 'virtual' member to each staff's care team.

Share this project: