Deploying medical chatbots in low-resource, multilingual settings requires multiple layers of safety (triage, retrieval, generation) and multi-method evaluation—no single model or test is sufficient for trustworthy healthcare AI.
Researchers built a phone-based chatbot to answer maternal health questions in India, where users often have limited health literacy and speak multiple languages. The system combines triage (routing urgent cases to experts), retrieval of curated health guidelines, and AI-generated responses.