MAIA — Maternal AI Assistant

Inspiration

Pregnancy and postpartum recovery can be overwhelming, especially for women who do not always have immediate access to medical guidance or emotional support. Many mothers experience anxiety, confusion about symptoms, and uncertainty about when to seek professional help.

Our team wanted to create an AI assistant that acts as a supportive first layer of guidance for maternal health. The goal was not to replace doctors, but to help mothers quickly get helpful information, reassurance, and guidance when they need it.

This led us to build MAIA (Maternal AI Assistant) — an AI-powered voice and chat assistant designed to support pregnant and postpartum women with safe, empathetic responses and guidance.

What it does

MAIA is an AI Doula assistant that provides:

Conversational maternal health support
Voice interaction using browser speech recognition
AI-generated responses for pregnancy and postpartum questions
Safety-first responses that encourage professional medical help when needed
A clean and simple interface designed for accessibility and ease of use

Users can either type their questions or speak directly to MAIA, making the interaction feel natural and supportive.

How we built it

The application is built as a modern web application with an AI-powered conversational layer.

Core components include:

React + Vite for the frontend interface
TailwindCSS for UI styling
Groq LLM API for fast AI responses
Web Speech API for voice input and speech synthesis
Firebase Hosting for deployment

The AI assistant uses a structured system prompt designed specifically for maternal care scenarios. This prompt ensures the AI:

avoids diagnosing medical conditions
provides calm and supportive guidance
recommends seeking professional help in emergencies

The app supports both text chat and voice interaction, making it easier for users to communicate naturally.

Challenges we ran into

We encountered several interesting technical challenges during development:

Voice recognition issues

Browser speech recognition APIs behave differently across environments. We had to handle microphone permissions, browser compatibility, and recognition errors.

Preventing multiple AI requests

Speech recognition events sometimes trigger multiple results while a user is speaking. We implemented logic to only send final speech transcripts to the AI model.

API limits and reliability

We initially explored different AI providers and had to handle rate limits and fallback error handling to ensure the UI never froze during API responses.