Sakhi Chatbot

Every day, more than 300 million women are going through their menstrual cycle. Many women and girls do not have the right information due to constrained education and the stigma around this subject. Only 6% of girls in Bangladesh have a menstrual hygiene management (MHM) education and 53% of adolescent girls know about menstruation before their first period. 

Around 41% of Bangladeshi girls take an absence from their schools during their period, missing 2.5 days a month, on average. The popular WASH literacy-focused physical intervention succeeded in improving MHM education for women and girls in Bangladesh. However, it is extremely challenging to scale such physical interventions to large populations without significant investment of time and resources. 

Many similar prior attempts have pursued physical deployment including school, peer, or healthcare provider-led camps, information kits, and education to influence psychosocial factors, and address harmful taboos and stigma. They often struggle to reach remote areas and may lack interactivity. Earlier digital solutions, such as mobile apps and websites, provide accessibility but may require internet connectivity, excluding marginalized communities. 

We leverage large language models (LLMs) to scale literacy interventions and improve awareness of menstrual health and hygiene among these young girls and women.

Product demo: https://sakhi.simppl.org

We present a Bengali WhatsApp chatbot to deliver a digital literacy intervention for improving menstrual hygiene management awareness among local populations in Bangladesh. We use large language models grounded in verified and accurate information to support community-based participatory research in partnership with local nonprofits and public health centers We have won a few grants and demonstrated the viability of our prototype. We are deploying pilots and the converting our early-stage partners into long-term partners to deploy monitoring and evaluation for a months-long digital literacy program.

Alongside a local partner, we propose a WhatsApp chatbot that generates responses grounded in accurate, verified knowledge from international health agencies.

We have created a chatbot that responds in Indic languages such as Bangla, thus making it accessible for the low-resource regions of Bangladesh. It uses WhatsApp, a platform used by 44 million Bangladeshi users to interact with the chatbot [1]. By providing accurate information about menstrual health, it fosters a sense of safety amongst young girls and women to openly ask their questions without feeling insecure. We have designed our system with safety measures such as rate limiting, content filtering, and data privacy protections. By curating verified content from trustworthy sources, our system enables users to learn interactively at their own pace about sensitive topics like health and hygiene, while minimizing risks from misinformation.

[1] -https://worldpopulationreview.com/country-rankings/whatsapp-users-by-country

There are serious barriers affecting menstrual health literacy in bangladesh for young girls and women who face negative health outcomes and systematic disadvantage for educational opportunities arising from compounding health and psychological side-effects.

Traditional approaches like educational workshops and printed materials often struggle to reach remote areas and face-to-face sessions may perpetuate societal taboos. Moreover, printed materials may lack interactivity. Digital solutions, such as mobile apps and websites, provide accessibility but may require internet connectivity, excluding marginalized communities. Some initiatives rely on community health workers, but resource constraints limit their reach. Our community partners say past efforts may have faltered due to insufficient scalability, cultural insensitivity, or a lack of technological integration. The proposed WhatsApp chatbot leverages existing communication channels, ensuring widespread accessibility. Our solution overcomes these barriers by combining language models, local partnerships, and user-friendly platforms for effective and culturally sensitive menstrual health education. 

We are members belonging to communities affected by the social stigma and taboos in India and Bangladesh that came together as part of a research collective called SimPPL (https://simppl.org) and built Sakhi our chatbot to address problems close to home, based on our conversations with nonprofits working in public health, fact-checkers, investigative journalists, and digital media orgs. interested in online information sharing. Our motivation stems from recognizing the urgent need to address menstrual health challenges affecting millions of women and families with complex indirect effects on wellbeing. At SimPPL, before building Sakhi, we have worked with the Sunday Times, Deutsche Welle, Ippen Digital, New York Public Radio, Aadhar Bahuddeshiya Sanstha, Chambal Media, Jagran New Media, Lions Club, Heal Station Foundation, DigiSwasthya, and others collectively serving millions of consumers with both accurate and verified information as well as running nation-scale physical and digital public health programs.

Our team comprises members who are data scientists, ML engineers, political scientists, and social science researchers well-equipped to build scalable, interpretable ML tools including those with cutting-edge research experience with LLMs, publishing at EMNLP, ICML, NeurIPS, ICON, ACL, and NAACL. We have a combination of practitioners, researchers, and locals who are well aware of the problem and can articulate well what potential solutions might look like.

Swapneel Mehta: I am a postdoctoral researcher investigating platform governance and free speech at Boston University and MIT. I hold a Ph.D. from NYU's Center for Data Science, specializing in machine learning, causal inference, and their applications in combating online misinformation. I founded and lead SimPPL, a nonprofit research collective with the mission of improving trust on the social internet. We train undergraduate students to build responsible AI tools at scale, winning grants and awards from Google, Wikimedia, Amazon, Mozilla, and others. In a past life, I used to work on machine learning at Twitter, Slack, Adobe, Oxford, some startups, and CERN.

Smriti Bhaya: I am a Master's in City Planning Candidate at MIT, focused on resiliency planning and risk mitigation that is adaptable and equitable. My interests also lie in social entrepreneurship and innovation in the built environment. To this end through Purifyx, we are working to provide clean water solutions in rural communities of India with low-tech, low-cost passive technologies. With a background in architecture, I have worked on architectural, urban design and policy projects in India, on a range of public and private projects.

Arpa Paul: I am a Bangladeshi-American healthcare professional with over a decade of work experience with underserved and indigenous communities in Queens, New York. I am a Skidmore alum with a B.A. in Psychology and lead the operations at five locations for New York Opthalmology. I am deeply motivated to provide a higher quality of direct patient care and clinical psychology research.

We have launched pilots with Spreeha Foundation, launching Sakhi to over 200 users in 6 public health centers around Dhaka and with the Marma community in the Chittagong health tracts launching a pilot to over 100 young girls in partnership with staff at the Asian University for Women. Additionally we have secured a Memorandum of understanding with the Aadhar Bahuddeshiya Sanstha in India operating healthcare centers in over a hundred villages in Maharashtra, and secured letters of commitment from three other nonprofits in India working on providing digital healthcare solutions. We have two commercial partnerships, one deployed for Davinci Wearables led by former MIT founders, and are in an early conversation with SafePad Bangladesh as well as receiving an invitation to present to the World Bank's eMBeD unit that works on behavioral interventions globally.

Our participation in social innovation programs including the IDEAS Social Innovation Challenge has encouraged us to push the envelope with our solution, fostering creativity and innovation through the conversations we strike up within our cohort and with mentors. This has proved invaluable, and directly resulted in the partnerships we listed above. Beyond the product growth and development we aim to achieve, we are excited to be part of a community of driven individuals that have a deep understanding of healthcare and educational challenges in addition to the technological knowhow to tackle them. Human behavior is often unexpectedly rigid at times, and malleable at other times, and almost always, effective interventions demand empathy, creativity, and resilience to deploy. As a small example, we hope to be able to gain insights from the cohort on how their respective communities have responded to behavioral interventions in the past and what kind of efforts can support future delivery of digital interventions that we are hoping to deploy globally. Our participation helps us to maintain a razor sharp on the changemakers we hope to become for communities that we have grown up in.

Every day, more than 300 million women are going through their menstrual cycle. Many women and girls do not have the right information due to constrained education and the stigma around this subject. Only 6% of girls in Bangladesh have a menstrual hygiene management (MHM) education and 53% of adolescent girls know about menstruation before their first period. 

Around 41% of Bangladeshi girls take an absence from their schools during their period, missing 2.5 days a month, on average. The popular WASH literacy-focused physical intervention succeeded in improving MHM education for women and girls in Bangladesh. However, it is extremely challenging to scale such physical interventions to large populations without significant investment of time and resources. 

Many similar prior attempts have pursued physical deployment including school, peer, or healthcare provider-led camps, information kits, and education to influence psychosocial factors, and address harmful taboos and stigma. They often struggle to reach remote areas and may lack interactivity. Earlier digital solutions, such as mobile apps and websites, provide accessibility but may require internet connectivity, excluding marginalized communities.

Sakhi advances the use of AI for social good through our chatbot designed to intervene on the lack of knowledge about menstrual health and hygiene among young girls and women who are systematically disadvantaged as a result of their health issues. Uniquely, we capitalize on the low-barrier to entry provided by WhatsApp and the trust built by our local partners running decades-long health programs in the community. We leverage a custom dataset curated in Bangla and English to improve the chatbot's accuracy. Our team's domain expertise working with local stakeholders on healthcare services and building the largest language models in India helps create a unique technological edge for our solution including the ability to converse and share images and documentation with the chatbot. Most importantly, we build a robust monitoring and evaluation system to measure the interventional effects through a novel research design supported by leading experts in psychology and human behavior at MIT and the World Bank.

There is a history of successes of healthcare interventions in Bangladesh termed as WASH (

Our theory of change is informed by research and user interviews conducted among our target audience in Bangladesh and India and additionally through research our team has conducted globally deploying interventions in low and middle income countries with our partners.

We aim to improve menstrual hygiene awareness among rural populations by at least 15% and urban populations by at least 25% within a year of deploying our intervention, at a cost that is a fraction of the estimated costs for deploying health and sanitation awareness interventions. Locally, we measure our impact through the number of local operational units that are equipped and able to deploy and monitor digital literacy interventions for improving women's health outcomes tying with SDG indicators for gender equality and sanitation. Sakhi's performance and efficacy is measured through in-app conversational metrics that highlight the fulfilment of customer queries, diversity and frequency of incoming queries, and degree of shift towards informed personas that we have developed measurement models for using natural language processing (NLP) technology. For local projects, we partner with nonprofits to run pre and posttreatment surveys to measure the individual differences in behavior and knowledge caused by our digital literacy intervention. We employ this survey-based research design in staggered waves to measure the lasting change produced up to a few months after the deployment of our system. Finally, we partner with providers of commercial products to measure product uptake metrics to monitor the regional participation in potential programs subsidizing the provision of menstrual hygiene products to local populations informing efforts by governmental and nongovernmental agencies. At present we have three partnerships that we have received interest into converting into a year-long program in Bangladesh and scheduled presentations to international agencies who wish to discuss a state-wide deployment, making significant headway towards our goals!

We use multimodal AI chatbots deployed on WhatsApp to power our solution. We have built a custom retrieval augmented generation pipeline distributed over a load balancer serving fine-tuned large language models (LLMs) in a plug-and-play fashion to efficiently scale our system to thousands of users each day. This system, complemented by our monitoring and evaluation platform and engineering optimizations, means that we can offer our solution at highly competitive pricing making it an attractive proposition to programs in the global south that typically cost a few dollars a day per person, for health and sanitation interventions, including those run by UNICEF in Bangladesh. Key to our solution is our domain knowledge and human experts that inform the data curation of the training materials for our LLMs which, research has shown, increases the quality of the output by orders of magnitude. In order to include those that don't type we enable audio-based conversations for our intervention to scale beyond digitally literate communities. Over 44 million Bangladeshis use WhatsApp, and we are already running pilots at multiple health centers for urban slum-dwellers in Dhaka and with rural populations in the Chittagong health tracts. From a commercial, our monitoring and evaluation platform works in tandem with nonprofit interventional monitoring and for menstrual health products, receiving inbound partnership requests from reusable menstrual pad providers to better connect them with their audiences.

Full time staff: 6 members; CEO, CTO, COO, Research Lead, Technical Leads

Part-time staff: 4 members; Program Managers, Operations Leads

Contractors: 10 members; We train students in India and Bangladesh as developers and researchers to advance our understanding of digital literacy interventions through our nonprofit, SimPPL.

We have built Sakhi for 9 months, since September 2023. SimPPL, the parent nonprofit out of which Sakhi was born, has been in operation since 2021. We have received support through grants from the MIT PKG IDEAS Social Innovation Challenge. The technology that goes into Sakhi was developed over a year including a foray into WhatsApp data collection, web, and Telegram chatbots, before identifying the correct product-market fit for our target audience.