Virtual Mentor

A wearable, voice-activated AI chatbot that guides health workers through treatment protocols

We’re building the first wearable, voice-activated AI chatbot that will guide health care workers through the assessments and actions indicated in a time-bound clinical scenario. While we think this “Virtual Mentor” will have varied applications, we are particularly interested in maternal and newborn survival in low- and middle-income countries (LMICs).

The largest proportion of maternal deaths in LMICs—about 100,000 deaths per year—is due to postpartum hemorrhage (PPH), heavy maternal bleeding after birth. 90% of global deaths due to PPH happen in LMICs. When treated in a timely manner, post-partum hemorrhage can be addressed, but if left untreated or treated improperly—like in this case—it can lead to the death of the mother.

The good news is that more women in LMICs are giving birth in hospitals and other frontline healthcare facilities, rather than in locations ill-equipped to handle emergencies. The bad news is that there are not enough skilled birth attendants to handle the new demand. What’s more, many attendants don’t have enough knowledge about how to treat emergencies like postpartum hemorrhage.

 A 2018 report on maternal death published by the government of Kenya found that in 91 percent of cases, “a different management of care could have made a difference to the outcome.” We must equip birth attendants not only with the training, equipment and technology to care for patients, but with the right kind of support to give them the confidence to save lives. Among that report’s recommendations was “embrace and scale up the use of technology to enhance access and availability of quality care.”

We were awarded seed grants by Saving Lives at Birth and the Johnson & Johnson Gen H Challenge to build and test the Virtual Mentor prototype, but we need additional financing for the development team required to test the Virtual Mentor between October 2018 to March 2019. We have built the current version of the Virtual Mentor using several components of the IBM Watson platform. We primarily use “AI” for two purposes: 1) to accurately understand what is said, and 2) allow the Virtual Mentor to be more flexible and forgiving.

We have iterated the Virtual Mentor prototype to find common pitfalls like differences in timing, performance, and step order. We use this data with Watson’s AI platform to make the system much more flexible with input and much more forgiving with skipped steps or “incorrect” responses. We use another tool, Snowboy, to train models for a unique use of wake words to reduce the overall dependence on connectivity and make the device more intuitive and responsive.

We want to augment the capacity of every health worker to adhere to treatment protocols in Kenya and other LMICs—treatments for obstetric hemorrhage and other time-sensitive, hands-on emergencies. We have tested the Virtual Mentor prototype with ideal users (nurses and midwives) in Kenya, and now we’re asking for Solve’s support to develop this wearable AI chatbot so we can pilot Virtual Mentor at 99% reliability.

Virtual Mentor proposes two hypotheses. First, an AI chatbot will augment health workers' cognitive and physical capacity to save lives when they need both hands and every second matters. Second, we can reduce the platform's dependence on cloud connectivity by training models with machine learning methods.

Our focus now is developing a prototype to continue to test and refine
Virtual Mentor. In March-April 2019 we want to pilot a 99% reliable Virtual Mentor in Kenya, among nurses and midwives, during highly-realistic simulated emergency drills. By July 1, 2019 we will expand the clinical scenarios for which the chatbot can assist, spread to training at more facilities in Kenya, and plan for spread to other LMIC contexts.

It is our vision to scale Virtual Mentor to other emergency scenarios, such as neonatal resuscitation and preeclampsia, and to put the chatbot platform on a wearable device with which users (health care workers) engage frequently. We will work to provide Virtual Mentor to frontline health workers in Kenya and other LMICs, along with the training and support this roll out will require. We will build analytics tools into the Virtual Mentor platform to monitor use and iterate algorithms with the aim to help health workers optimize their performance during life-threatening complications.

We are working on our strategy and defining our position in the value chain. Our current hypothesis is that health facility or health system directors will purchase Virtual Mentor for use by birth attendants in maternity units.

Virtual Mentor is in the prototype stage. We do not have customers yet.

Over the next 12 months we will continue co-designing Virtual Mentor with nurses and midwives in Kenya, and will pilot-test it with at least 20 of them. Virtual Mentor will start as a training tool to practice correct management practices in simulated emergency drills, and we want to work with birth attendants in 30 facilities in Kenya over the next 3 years. We are actively looking for partners with whom to test and spread Virtual Mentor to other contexts, once we achieve 99% tech reliability.

Our team consists of a midwife who understands the work processes of a frontline health worker, a Kenya obstetrician-gynecologist with the desire to mentor every birth attendant in sub-Saharan Africa, a US-based obstetrician-gynecologist with 20 years of global health experience, a product designer with extensive experience owning AI-assisted projects, and a maternal-newborn health fundraising and project management strategist.

We currently manage and distribute our grant funds through the University of California, San Francisco Institute for Global Health Sciences while we develop and iterate the Virtual Mentor prototype. Once the prototype is optimized as a minimal viable product, we plan to start a for-profit company. Our target customers will be facility and system directors in LMIC contexts with discretionary budgets available and interest in tech-assisted decision support.

The Virtual Mentor team does not yet include a software developer, and we can't move forward without either funds to hire a development team or with the participation of a developer who is interested and available. We hope that Solve can help us connect with a developer (or developers) whose time is supported by another organization.

Our team has deep understanding of the needs of front line birth attendants, the context of their daily work, and the stubborn conviction that voice-activated technology should be leveraged to assist them. However, we don't have software development expertise and this is a fundamental barrier to our progress. We hope that Solve will help Virtual Mentor connect to potential development partners, as this fundamental need is for labor that is expensive and we do not yet have the budget to hire excellent developers at the market rate.