Rementia

A HIPAA-oriented dementia care assistant with real-time voice and text chat, RAG over medical knowledge, OCR prescription extraction with drug-interaction checks, and caregiver dashboards so context is shared safely.

Role. Built patient and caregiver experiences plus supporting FastAPI services for chat, voice, retrieval, and encrypted storage.

Healthcare
Privacy
React
FastAPI
Bedrock
RAG

Problem

Tools that only chat with the patient leave caregivers in the dark. Prescriptions arrive as scans. Health data needs encryption at rest, not demo shortcuts, and voice has to work in the same loop as text.

Tradeoff

A narrower feature set outside the care loop so privacy, voice, and retrieval stay correct first.

What we considered

Text-only first, caregivers and patients often need voice.
A third-party chatbot, weak control over encryption, retention, and clinical RAG.
Separate caregiver tools, context stays trapped in the patient thread.

How it works

React 19 web app for patients and caregivers. FastAPI backend with PostgreSQL, JWT auth, and Fernet encryption for sensitive fields. Real-time voice/text over WebSockets. RAG over medical knowledge via AWS Bedrock (Nova Sonic, Titan). OCR extracts prescriptions and checks interactions. Caregiver dashboards share context without exposing raw secrets. Deployed with Docker and Nginx.

Decisions

Caregiver dashboard and patient chat are one system.
Encrypt sensitive fields at rest; do not rely on network isolation alone.
Treat voice as a first-class channel beside text.

Outcomes

Realtime voice and text with retrieval and prescription help in a privacy-first setup.
Caregivers see shared context without living inside the patient transcript.
Hardened deploy path with Docker and Nginx.

Want to go deeper on this project?

Happy to walk through decisions and constraints.

Ask about this project