"Kya family planning after marriage hoti hai?": Integrating Cultural Sensitivity in an LLM Chatbot for Reproductive Health

TL;DR

The paper investigates how to design LLM-based chatbots for sexual and reproductive health in culturally diverse, underserved settings. It uses a Mumbai-based collaboration with Myna Mahila Foundation and a RAG-enabled GPT-4 system deployed in Hinglish, evaluating 2118 user–bot interactions plus qualitative stakeholder input. The authors identify strengths in capturing local language, social norms, and preferences, but also highlight gaps in handling taboos, legal issues, and socio-economic constraints, proposing a four-layer context framework to guide design. The work contributes a scalable, community-driven methodology for building culturally sensitive LLM interventions in public health, with practical guidance for balancing medical accuracy and cultural relevance in low-resource environments.

Abstract

Access to sexual and reproductive health information remains a challenge in many communities globally, due to cultural taboos and limited availability of healthcare providers. Public health organizations are increasingly turning to Large Language Models (LLMs) to improve access to timely and personalized information. However, recent HCI scholarship indicates that significant challenges remain in incorporating context awareness and mitigating bias in LLMs. In this paper, we study the development of a culturally-appropriate LLM-based chatbot for reproductive health with underserved women in urban India. Through user interactions, focus groups, and interviews with multiple stakeholders, we examine the chatbot's response to sensitive and highly contextual queries on reproductive health. Our findings reveal strengths and limitations of the system in capturing local context, and complexities around what constitutes "culture". Finally, we discuss how local context might be better integrated, and present a framework to inform the design of culturally-sensitive chatbots for community health.

Figures (5)

• Figure 1: Various phases of the study.
• Figure 2: Chatbot Interface on Mobile Web Platform. The figure presents a greeting in Hinglish and an explanation of what the chatbot is capable of. It also presents buttons below the text with suggested questions to click on to get started. On the right is a button with the speaker icon which plays automated text-to-speech on clicking.
• Figure 3: Final system architecture of the chatbot. The three stages of the chatbot flow include---Translation module, Generating the medical answer, and Localization module. The translation module involves OpenAI’s LLM model which interprets and translates the user’s query in Hinglish into English. Generating the medical answer involves generating the medical answer from the chatbot’s knowledge base by prompting the LLM with a predefined prompt and translating it back to the user language. The localization module involves replacing complex medical words with colloquial terms.
• Figure 4: Timing of questions asked by users. The graph summarizes the number of queries received at different times of the day. The highest activity was between 5AM and 12PM.
• Figure 5: The four context layers that shape cultural relevance of LLM-generated text.