An LLM voice agent can map unconstrained multilingual free-text survey responses to structured SDoH answers with high accuracy

A voice-based conversational agent that transcribes (and, for non-English speech, translates) spoken answers and passes them to a large language model can map unconstrained free-text survey responses to the correct discrete structured answer choices with high accuracy across more than one language, offering a language-agnostic, interviewer-free way to collect SDoH data from patients including those with limited English proficiency.