MaternaAI: A Multimodal AI Agent for Pregnancy Symptom Guidance

Abstract

Pregnancy related complications are one of the top causes of morbidity for both mother and child worldwide. In low-and middle-income countries, where access to antenatal care is limited, timely and accessible symptom triage is especially important. This paper presents MaternaAI, a Multimodal intelligent pregnancy companion using a combination of three different artificial intelligence technologies: “Retrieval Augmented Generation (RAG)” over the ACOG clinical knowledge base for well-informed symptom triage; “ensemble machine learning-based scoring models “using Gradient Boosting for postpartum depression, Random Forest for preterm birth and preeclampsia, and a CDC PRAMS data set of 5,000 records; and Gemini 3.1 Pro,” A multi-turn reasoning LLM. In terms of performance, when evaluated using a held-out set of 1,000 records for each model, MaternaAI achieves an ROC-AUC of 0.662 for PPD risk, 0.577 for preterm birth, and 0.615 for preeclampsia. The system also includes a Cardiff-method-based kick counter, Edinburgh Postnatal Depression Scale-based screening, contraction timer, due date calculator, and trimester-based nutrition guide, all delivered as a React.js front end with a FastAPI back end. To the best of our knowledge, MaternaAI is the first system to integrate RAG, multi-model ML-based risk scoring, and a large language model into a single perinatal health agent.