Kadam: Increasing Access to Prenatal Care in Developing Regions

Abstract

In developing economies, access to healthcare is defined by financial accessibility, geographical availability, and community acceptability. In remote areas of low-income countries, access to healthcare is even more limited. Medical malpractice and unregulated practices in these communities endanger human lives. The danger to human lives becomes even more severe in applications of prenatal care. Compared to the developed world, only a limited percentage of women have access to prenatal care in low-income countries. The percentage is even lower among the rural population of these countries. Given this, the goal of this thesis is to increase access to prenatal advice in remote regions of developing countries to reduce perinatal mortality.

This thesis details the development and evaluation of Kadam, an offline, low-cost, and deployable conversational chatbot. Kadam provides pregnant women with necessary information regarding pregnancy to keep them informed. We first create a pregnancy data set by scraping WebMD. Using our data set, we build Kadam through Rasa which is an open source framework that allows third party developers to build AI-based contextual assistants. We train Kadam on Mac OS and afterwards deploy it on a low-cost hardware, Raspberry Pi Model 4 B. Following the evaluation, we found that response time of Kadam remained consistent between high-computing device like a Macbook computer and low-computing device like a Raspberry Pi device. Kadam can be improved with training on more robust, real-life data in multiple languages.