Our recent survey pointed to the lack of sufficient menstrual help and awareness on campus. Navigating the alley of menstruation may feel like uncharted territory, but there’s a bright light at the end: ISHA (Intelligent System for Healthcare Assistance).
Created by Prof. Tanmoy Chakraborty and his research team, ISHA is an AI Companion designed to answer a wide range of menstruation-related queries with accuracy and reliability. Recognising the potential of this innovation, we interviewed Prof. Chakraborty to gain invaluable insights into the creation of ISHA.
Prasoon: What was your inspiration for ISHA?
Professor: Our team focuses on natural language processing, text, and speech processing with a specialty in large language models and generative AI, particularly in mental health applications. Over the past 4-5 years, we've developed AI tools for mental health counselling. Recently, we've been exploring AI models tailored for specific uses. Recognizing the stigma around mental health and women's health in India, especially menstrual health, we are motivated to challenge these stigmas. Our idea involved creating an Intelligent Healthcare Assistant to support women by providing suggestions and information about menstrual health, empowering them to discuss and manage their health issues openly.
Akshit: What difficulties did your team face while working on ISHA?
Professor: Foremost, we had trouble scrounging relevant data because of the relatively small number of documents related to menstrual health. Additionally, the data was redundant, so collecting appropriate data to fine-tune such a gigantic model took a lot of work, which forced us to generate data and interview different health workers synthetically. Evaluating the system was another challenge for us as we are not mental health experts or gynaecologists, so we had to depend heavily on doctors. We consulted doctors at AIIMS and female students at IIT Delhi for evaluation. Ensuring the model followed LLM guidelines and had proper guardrails to prevent producing vague responses was crucial. Making it fast and scalable was another challenge.
Prasoon: What sources of funding and support have been crucial for the development of ISHA?
Professor: I have not approached funding agencies because I wanted a prototype ready first. My other unrestricted awards fund students and RAs working on this project.
Akshit: How is ISHA better than its competitors?
Professor: As far as I know, there is no dedicated AI based chatbot for menstrual health. ISHA represents one application of the fine-tuned LLaMA model we've developed. We plan to make the fine-tuned LLaMA model publicly available soon. While other tools like ChatGPT and Gemini exist, they often give vague and lengthy answers. We wanted our model to provide precise, concise suggestions for menstrual health problems.
Prasoon: Which community groups have you explicitly engaged for this project?
Professor: So far, we have mostly interacted with students at IIT Delhi and other colleges and hospitals. We require more responses, especially from the female community, including those in rural and economically backward areas. Currently, the tool is web-based and dependent on the IIT server to serve the public users. We plan to design an app-based system as we get more funding.
Prasoon: How do you think people's perspective about the menstrual cycle would change through ISHA, especially regarding the taboos around it?
Professor: While discussing this with my wife and mother, we found that taboos exist even in urban society. If both men and women start using this tool, they will understand and hopefully break these taboos. The tool can suggest what to do at certain times, like handling pain or irregular cycles, without prescribing medication.
Prasoon: What potential challenges await ISHA, and how do you plan to tackle them?
Professor: As ISHA is to cater to a significant rural population, the language barrier is a primary concern. The tool is good in English but needs to be more effective in regional languages. We plan to incorporate regional languages and a speech-based understanding system.
Akshit: What advice do you have for students aspiring to work in this industry?
Professor: I suggest dedicating yourself to creating something positive for society using available resources. Conduct research, develop prototypes, and put them into action. Concentrate on urgent issues such as healthcare, environmental pollution, and agriculture. Begin working on initiatives early, even as early as the second year of college, and seek advice from mentors and faculty.
We also interviewed two motivated research students who worked on ISHA, a confidential and culturally sensitive platform for spreading awareness and knowledge. From a humanities background, one of the students was responsible for verifying the chatbot's answers, coordinating with gynaecologists and public health workers, and ensuring the responses were accurate, relevant, and sensitive to the Indian context. One key challenge was collecting and processing around 21,000 questions and answers over three months and then training the model to align with Indian cultural norms. The chatbot needed to handle sensitive topics delicately, such as addressing beliefs about menstruating women entering temples without judgement. An example highlighted the chatbot's ability to respond from a mother's perspective, showing the depth of cultural integration achieved. Questions about entering the kitchen or sitting at the family table during menstruation underscored the deep-rooted taboos and the necessity of ISHA.
To gauge the curiosity of the student community about menstrual health, we asked them for their queries regarding menstruation; here are some questions that caught our attention.
Menstrual awareness is unfortunately a huge issue in our country and this chatbot is a step towards combating prevailing misinformation, ignorance and taboos. ISHA, the menstrual health care assistant, helps in building a truly equitable society with its balanced and user-friendly answers. Use the link below to give it a try. Your feedback would be truly valuable in improving it!
Report by: Akshit Abhinav Kujur, Prasoon Raj
Design by: Himesh Rustagi, Nishka Muraka, Tejas Tyagi
Comments