RAAJI

Due to lack of education & healthcare infrastructure in most developing countries (e.g Pakistan), girls lack knowledge about sexual health, contraception & menstrual hygiene management practices. As a result, they develop infectious diseases, fertility issues and deal with unintended pregnancies that result in unsafe abortions. This becomes a vicious cycle leading to cities becoming overpopulated and citizens becoming unhealthy.

Our solution is an AI-powered chatbot called Raaji to equip girls across developing countries with knowledge to understand effective birth control methods, improve menstrual health management, & educate them on STDs. Raaji can also provide access to proximal health care resources or help users get connected to human support networks in cases of emergencies. 

At scale, Raaji can positively impact the health of millions of girls in many countries as her content can be contextualized for multiple languages and can be connected to SMS in cases of internet unavailability.  

In the last few decades, the world has seen exponential increase in human population. However, this spike has been most apparent in developing countries. This is because there is limited access to family planning information, education and services available. Due to this, girls and women bear the brunt of early/multiple pregnancies, infectious diseases, unsafe abortions, malnutrition, poor sanitation & hygiene, increased violence and harassment.

In a developing country like Pakistan, recent studies conclude:

• 49% had no knowledge of menstruation prior to first period. 79% of Pakistani women are not properly managing their menstrual hygiene. Incorrect menstrual waste disposal methods are leading to negative impact on the environment. 1 out of 3 girls miss school days every month during period. 

• Pakistan has low contraception usage and the world's highest abortion rates.

• Myths, stigma and shame make it impossible to seek help for sexually transmitted diseases.

How can cities and communities be happy, healthy and equal when half of its population continues to feel dis empowered? 

We believe that for healthy and well-functioning cities, we need empowered girls & women who have information to reproductive health education and access to services.   

In urban areas of developing countries, a large part of the population lives in slums. Women and girls who live in these slums are disproportionately underprivileged and do not have access to education, healthcare, employment, mobility and financial resources. If we educate these girls from an early age about their reproductive health, it would lead to improved quality of life in slums and can help the government achieve SDG goals & decrease dependency on healthcare system. One case we have worked with in urban slums is:

"12-year-old Yusra lives in an urban slum ‘Neelum Colony’ in Karachi, Pakistan. She attends The Garage School. Speaks/writes in Urdu. Hesitant to inquire abt health topics. Does not own a mobile but often plays with elder sibling’s phone."

Before our visit:

1. 40% of girls thought periods are something shameful, 25% thought that it isn’t whereas 35% couldn't decide between either.  

2. 75% of the girls knew the right duration of periods. 

3. 40% of girls believe they could use cloth or cotton during periods. 

Once they interacted with Raaji, the results showed:

• Majority of audience stopped considering periods shameful, knew the right duration, products, disposal techniques, cravings and pain associated with it. 

Detailed report here.

We are a Karachi-based startup & we are using AI to help address the societal challenge of the widespread unhygienic reproductive health practices.

Lack of education on these topics in developing countries is leading to girls developing fungal infections, reproductive tract infections and making them vulnerable to infertility.

Our solution is Raaji, a chatbot which uses Artificial Intelligence and Human expertise to answer questions around reproductive health. Through Raaji we provide girls with easy information accessible through a single application. Through Raaji we intend to give easy accessibility to knowledge, professionals and products.

In cases of emergency or where more empathy is required, we have designed an easy way for doctors, psychologists and lady health workers to take over a conversation and solve the concern of the user. In the future, we will use their interactions to train our model further. We are also working on integrating e-commerce players where we can help users to order different health products while endorsing these brands in terms of ad placement on our platform.

We provide user anonymity and do not capture email or phone numbers or social sign-ins. In the future we plan to take information such as age, gender and location. This will help us analyse our data and provide a more structured approach to our interactive education through this chatbot.  

• We use (supervised) machine learning for the natural language understanding (NLU) model which drives our “intent classification” i.e. the intent of the user. The training data is created by our team, and we train, test and optimize the NLU model

• We also use machine learning  to train a neural network model for conversational dialog. We trained this model using interactive reinforcement learning.

• We use data science for advanced analytics on the conversational data that we collect - for eg. topic modelling, path analysis, clustering, and exploratory analysis in context of other datasets

Here is the tech stack

• BotSociety →  for conversation design and preliminary data labelling

• Articulate → for labelling data, creating branching structure and training NLU model

• Rasa → for interactive training of the core dialog model

• Python → for webhooks

• Docker for containerization

• Data Layer → S3, MongoDB, Redis, Apache Spark, SQL

• React → for the web version

Interact with our MHM web demo here

first version app on the store

Conversational flowcharting here

Appendix

Photos and videos

Press: https://www.auratraaj.co/press

Use cases for RAAJI:

In Pakistan, we see traditional approaches to information dissemination (i.e. non-digital, offline, & in-person) by international and local NGOs, private hospital networks (eg. AKU), FMCG’s (eg.Unilever), & institutional health programs (lady health workers, greenstar). Other approaches are: crisis-line & call centre (eg.Aman Telehealth), or an mHealth approach (eg. SehatKahani, RingMD).

These approaches are cost intensive, rely on human resources, prone to judgment & human error, do not iteratively improve using data, and not enough engaging for young users. They are also not scaleable. In terms of call centres, users are afraid of being tracked, recorded or their information at stake. Girls cannot take advantage of the health camps set up my mhealth startups as they can't take a consultation on their own. 

From our field work, we learnt that girls wanted to learn about their bodies, get access to resources in a private and non-judgemental space. A new approach to tackle these issues is to use AI-powered chatbots which provides information in a private and human like manner. There are international solutions (Eg: GirlEffect, Lily Health, & TIA) which are using this approach to tackle the issues mentioned above. 

Our bot also introduces complex topics via its animated videos and provides required information anonymously in local languages and allows a human takeover by gynecologists and psychologists. This extends empathy and professional advice so the user can talk to a human from the comfort of their home without revealing any personal information.

Our core technology is using Artificial Intelligence & Natural Language Processing to create a chatbot that provides empathy and information regarding reproductive health topics. Currently, our chatbot is built using Dialogflow which uses Machine Learning to classify user input and generate response accordingly. This requires us to create a training dataset for the model which is structured as: 

1. Intents -  training data to train a model and classify a users’ intent

2. Entities -  training data train a model to identify/disambiguate 

3. Responses - labelled training data that matches  a response to users classified intent

For the next iteration, we are intending to use the open source conversational AI framework Rasa to build our conversational chatbot. It provides us the benefit of customizing the model, create custom pipelines, build dialog model & interactively train the chatbot. In the Rasa framework, there are two components:

1. Core = a chatbot framework with machine learning-based dialogue management

2. NLU = a library for natural language understanding with intent classification and entity extraction

The centre of these models are the training datasets on which these models will be trained & they can be seen here.

In addition, we will use deep learning approach to build a speech recognition model and train it with the urdu language and other local dialect corpus to build a technology with native local language integration. More information on this module can be seen here.

Users can reach chatbot through an Android app and Web demo with a custom back-end that logs data and routes conversations to human experts.

We got to this solution from all the research, focus groups, interviews, and workshops we did with our core beneficiaries over a period of three years as Aurat Raaj.

We did not start off with a chatbot but realised its need after we started showing girls our cartoons in their classroom. Every time it ended, girls had lots of questions that we couldn't answer sustainably after we left. These arrived as phone calls, messages on social media and emails.  

When we started working on this solution, its need was validated by several organisations working on sustainable development goals. Such as WSA, Index, TVE Inspiring Change, SM4E 2018 and they also saw its need in other parts of the world.  

Our theory of change is demonstrated best here:

From the initial version of the app, we received these results:

Well defined data is very important to design best for the user. It also allows us to track the sensitivity of conversations based on the user’s background. However, uptil now the personal data we collected  was minimal to allow the user to feel safe. The data we have collected so far, includes, device ID of user. We collect the amount of emergency cases as well in our database.

We have also collected the number of conversations and which were failed by the chatbot and how many of them were actually resolved. This allows us to know the stats about the amount of human workforce we need. 

Our first iteration of the app has been downloaded 5000+ times, our web demo has collected over 10,000 interactions. 

From the animated content series we've screened digitally and offline, we've made it has received over 500,000 views: https://www.facebook.com/watch/?v=267267024093690

In the next 5 years, we will create an impact by reaching 1 million girls in Pakistan, and equipping them with knowledge converting them from uninformed / unhygienic practices to informed and hygienic, leading to girls back in school and women living in communities without shame. This will contribute to overall improved physical and mental health outcomes, as well economic productivity gains.

The snowball of this effect is better families, improved well-being and empowered women and girls in families and communities across Pakistan, who are motivated  to come out and talk about their problems even if they are taboo.

Our goals for the next year are:

1.Doing pilots, testing and collecting data to train the model.

2. Build webhook for integration with e-commerce provider.

3. Improve the local language pipeline technology integration.

4. Measure impact and test learning outcomes. 

Our goals for the next five years are:

We’ll prove that AI is mature to have knowledgeable conversations in important areas of society. 

In doing so, we potentially better the menstrual health practices of 5 million+ women around the world. With the use of formal monitoring and evaluation we would be able to prove the efficacy of our solution and help contribute to Pakistan sustainable development goals. Eventually, we expect to contribute significantly to improved knowledge, reduced stigma, improved health outcomes, and improved school attendance for girls in Pakistan. 

In terms of  technology/data,  we expect to have a robust/scalable technology solution (topics, geography, language). We hope to have developed multiple domain knowledge data sets, that we can use to do social good.  We have found a business model that enables us sustainability and growth. We hope to contribute to AI for good, open source, and nurture the open data ecosystem in Pakistan.

1) Adoption by end-users is a significant risk, especially as our target audience are largely not tech-savvy,  non-english speakers and cannot read or write/type. 

2) Potential of pressure from institutional and religio-political to stop awareness and outreach activities, limit ability to deploy technology to local networks and face scrutiny / forced shutdown

3) Our current design doesn’t provide a solution that supports speech recognition and text to speech conversion in local languages. Although technology solutions exist for speech recognition, the datasets needed for use with local languages in Pakistan are not easily accessible or they charge an amount not affordable to us (e.g., google translator)

Partnership with subject matter experts and groups that are experts in content/subject matter, generally do not have AI tools or technology, and have repeatedly expressed interest in working with us to convert their ‘traditional’ content into a chatbot that they can use in their outreach to create social impact.

1) We plan to address this by taking a user-centred approach to design the solution, build  iteratively and incorporate feedback/validation from end users to inform the design at each iteration. Additionally, we plan to address this by using voice, audio/visual content, and digital animation characters at the communication medium.

2) We plan to address this by forging collaborative partnerships with community leaders to understand concerns as well as health care networks and community support networks to make sure our messaging is culturally-sensitive.

3) We plan to collaborate with NLP & AI groups at local universities[27]  (including Government ICT grants and research centers) actively working for local language technology solutions to help us integrate above-mentioned solutions.

n/a

4 full time resources working right now.

Saba Khalid - CEO and Founder

Jaya Rajwani - Technology Lead

Sehar Palla - Marketing & Partnership Manager

Muqadar Ali - Software Engineer

2 part time team members

Ali Abbas - Design Lead

Suha Sulemani - Social Media & marketing

We have started running a fellowship program for University students. These students from top universities are learning tools we are working with, speeding our data collection, cleaning and AI training process.

Many incredible mentors devote their time: 

Ida Tin - Cofounder Clue

Christian Ehl - Digital The Do School

Valerie Mocker - Director Nesta

Jehan Ara - President of Pakistan Software Houses Association

The team is led by Saba Khalid who is an award-winning social entrepreneur, journalist, filmmaker, mentor, public speaker and blogger. She was the closing speaker for World Summit of Arts and Culture Kuala Lumpur in 2019. She won the She Loves Tech Pakistan edition in 2018. She has completed prestigious fellowships such as IFA Cross Culture, IVLP program, the Do School fellowship. She attended theTech Camp and won the Cultural Vistas grant to work with Indian health startups. She is a finalist for WWF Green Innovation, UNICEF MHM innovation challenge, Standard Chartered Women in Tech program. Currently, she is bringing The DO School's innovation hub Do-X to Karachi. 

Jaya Rajwani is technology specialist who has a deep interest in data science, machine Learning & AI for usage within civic innovation and social good. Jaya works on the chatbot to make it more human like and localize it to to be tested with young girls from urban slums.

Sehar Palla is a recent UCLA grad who has a keen interest in global studies, therapy and mental health. And helps us design empathy-driven content and helps us pilot it in schools. She also helps us reach more partners. 

Muqaddar Jamali is a recent computer science graduate with a year's experience working with an international software house He's willing to learn new technologies by choice and his exposure to different tools, technologies make him unique. 

We are all committed to social change in Pakistan and have always spent time volunteering for Pakistan's future. 

Pathfinder International

We've been in touch with their local as well as international office and they are keen to help us to develop our conversations and localised content on birth control methods 

Unicef Pakistan

We've been in touch with their local office but also their global innovation team and are exploring ways to pilot in UNICEF sponsored schools across Pakistan as well get more funding and research support. 

Santax

This local ecofriendly family owned sanitary napkin company has been donating pads for our pilots and has shared interest in receiving our insights from our pilots on the conditions for menstrual hygiene in Pakistan 

Non-profit Schools & private institutes in Karachi

Private institutes have shown interest in subscribing to Raaji's content so they can share it regularly through their labs as sex education course for kids going through puberty and adolenscence. While public school and non-profit schools have given us easy access to their morning and afternoon students. 

Other health startups

Other health startups have helped us design for various topics already. These include fertility startup from India called Infertility Dost and mental health platform Relive Now.

We plan to try different revenue streams:

- Schools subscribe to the full chatbot version and we charge them on a monthly basis.

- Brands for health are given key real time data that can help them target, market their products better. 

- Certain brands are endorsed within the chatbot's conversation for a certain time period. 

- We fulfill product orders for sanitary products through RAAJI. 

Started establishing partnerships with sanitary pad fulfillment partners for sponsored products. A local manufacturing company Santax is interested in selling through the platform as well as award winning sanitary napkin ecommerce app Girly Things from Pakistan.

Started building partnerships with NGO networks and other UN organizations who can benefit from this data. Government is also interested in this data as it helps them plan interventions.In case, these partners are not interested in buying the data, we would like to approach other organisations.

Already advertising random location based products through the app to show the organization how it will look. In conversation with health and hygiene companies for custom ads display.

Since a platform like this has never been set up before and it isn't easy to set it up in an developing startup ecosystem like Pakistan's & as a female-founded company, we are going to try various funding mediums to get to a point where can raise investment capital from impact ventures.

- Pitch competitions and prize money 

- Fellowships & grants

- Sponsored Pilot events in schools 

If we win an MIT prize, we hope to build an open source MVP and establisg our business model to monetize the technology and power the chatbot with affiliating marketing referrals and fulfillments (send traffic to health & hygiene brands) and start collecting data (for research & evaluation of outreach campaigns.

For inventors and innovators everywhere around the world, being able to go to MIT for Solve is a dream. This will allow us to connect with previous and current finalists working on similar tech or problems to gain a larger pool of knowledge. 

There is not enough funding for women entrepreneurs and especially those focused on femtech in Pakistan. Most funding goes to startups led by male teams working on problems that are profitable but not socially conscious. We want to be a role model for those innovating for social change using an emerging tech. This will not only give us a global platform but also inspire more solutions using AI and chatbots in the future. 

We would get a chance to learn and be mentored by experts in the MIT network. Find investors or grant organisations that are connected with our mission. 

We have limited access to AI experts in Pakistan, we might be able to find people willing to voluntarily contribute their knowledge time and resources 

We might partner with another startup team that can help us test our solution in another region or market or target audience and see if we need to pivot or grow or scale to another market or region right away 

Get more media coverage to make more connections that could be fruitful for our growth. 

n/a

• Our current design doesn’t provide a solution that supports speech recognition and text to speech conversion in local languages. Although technology solutions exist for speech recognition, the datasets needed for use with local languages in Pakistan are not easily accessible or they charge an amount not affordable to us (e.g., google translator). Therefore, we plan to collaborate with NLP & AI groups at local universities[27]  (including Government ICT grants and research centers) actively working for local language technology solutions to help us integrate above-mentioned solutions.
• Partnership with subject matter experts and groups that are experts in content/subject matter, generally do not have AI tools or technology, and have repeatedly expressed interest in working with us to convert their ‘traditional’ content into a chatbot that they can use in their outreach to create social impact.
• We want to find an acceleration program and we would love to see if there are possibilities to work with an MIT accelerator. We would love to see if we can partner with MIT students who can take up our project as their final year project. 
• We would love to meet with experts from RASA, DialogFlow, Bot Society, Google, Samsung. We could see if they could help us improve our design, be able to utilize their tech fully, and if they offer products to us voluntarily. For instance, we need testing devices but often don't have access to any.  

Our use of AI is currently limited to the underlying capability that is provided by DialogFlow, which enables us to build a deterministic flow based solution using decision making and NLP, but we are not using a full AI approach. We have a first version but lack a structured approach to model training and evaluation. 

The greatest deterrent has been that we realized that a) it is imperative that we build more validation and learn from hands-on testing about what type of conversations we need before trying to use AI, and  b) we did not have the funding to hire capability / talents to go beyond a deterministic flow-based solution. 

With the prize money, we can hire more talent and invest in more pilots so we can build an open source conversational AI framework. In addition, we will use deep learning approach to build a speech recognition model and train it with  urdu language and other local dialect corpus to build a technology with native local language integration.

We will establish baseline metrics for the performance of our models, and monitor these on on-going basis. Specifically, we will look at accuracy of intent classification, entity extraction and dialog prediction. We will measure the number of questions answered correctly, the number of topics that reach a certain level of correct answers, etc. to shows how AI learns and improves. 

Our work is based on human-centered design. We design with and for our end user. We constantly take the product to the market and keep an ear and eye for feedback. Our product can be used by schools, communities, urban slums on their own through an app or through one of our trained staff members. 

It also improves delivery of health information, coordination between various health networks and resources, provides ample data for better communication, and systems efficiency to achieve the goal of healthier cities. 

Our focus has been on a vulnerable developing country like Pakistan where infectious disease prevention is integral and potential for outbreak is tremendous.  

n/a

Our whole solution has been designed for the use of women by women. It is innovative, engaging and scalable. We will use the award by Vodafone Americas Foundation, we would explore potential for:

- Pilots in India as Vodafone has a huge presence in the country.

- We would take their advice and mentoring about in integrating our bot with SMS and the costs that could be involved using a service like Twilio. 

- Vodafone Germany has an accelerator called F Lane and we'd love to see if we could learn from some of the startups that have been through their program. 

Currently, we have a dataset which we are using to train a model that can do intent classification. We have collected data on topics and conversation logs that were asked by different user groups and the answers given regarding topics like menstrual health, harassment, online safety, abortion, birth control, etc.

Our newest tech team members (Zaki - https://www.linkedin.com/in/zakihpatel10/) and Jaya  https://www.linkedin.com/in/ja... ) have used data in the past for a range of applications. Both have technical capability in using data to build natural language processing, text classification, and information extraction pipelines and systems. 

For instance, Jaya worked on collecting and visualizing data for NGOs, NPOs and CSOs operating in Karachi to improve their outreach and connect with potential volunteers by analysing their social media profiles. She also analysed live RAM data to automate the digital forensics processes using Deep Learning.

We will use the prize money to make our dataset less biased towards English language and more inclusive of local languages and train our model to communicate with women from bottom of the pyramid. Moreover, the dataset we have is more based on our views of dealing with a problem, however, we have a team of volunteers to contribute to the content, but it still is not inclusive of views from our target audience. Therefore, we intend to counter this by conducting more field testing in future with the women and girls from different areas through local partnerships. 

Our team is led by women and designed for the use of women. By advancing their reproductive health rights, and giving them equal access to information, products and services at the tip of their finger, we are empowering them to be visible and have a say in society, make informed choices and take control of their bodies. 

If we win the award: 

We would make sure the human experts taking over sensitive and emergency conversations from the bot are a large number of trained lady health workers. These women usually have to go door-to-door to educate and provide products and their lives are always threatened by extremists. Our product would allow them to take action within safety of their own home and provide a closeby contact to a user going through any health or safety problem. 

Winning the prize would allow us to test in another UN market as well as with existing innovative UN solutions. One where this product could be integrated well and learn from is:

SIS BOT - Chatbot for abuse victims 

Our chatbot could also be introduced to Syrian refugees in the UN blockchain based cash for work program. 

Winning the prize might even create some interest from the Pakistan-based UN office. It will allow us to attend different global UN accelerators, labs or events and best explore and implement upon UN's AI policy around our user's data privacy, data protection and data ethics.