Memoral: Agentic AI for Dementia Care

Overview

Built in 24 hours at GreatUniHack 2025. Won three awards:

1st Place in the Reply Challenge: Best Use of AI Agents in Healthcare
3rd Overall out of 200+ participants
Hackathons UK Challenge Winner for Best Use of AI Agents on ARM

What Memoral Does

Memoral supports people living with dementia and Alzheimer's through:

Conversational engagement
Adaptive reminders
Intelligent task management

It improves daily structure, promotes cognitive engagement, and reduces caregiver burden without intrusive sensors or complex hardware.

For patients: Consistency and connection through a familiar AI companion.

For caregivers: Real-time visibility into well-being and daily routines.

Technical Implementation

Multi-Agent Architecture

Built a multi-agent system with specialized agents for:

Memory Agent: Manages patient context and conversation history
Task Agent: Handles reminders and daily scheduling
Health Agent: Monitors behavioral patterns and well-being indicators
Supervisor Agent: Coordinates agent interactions and escalations

Agent Orchestration

Used LangGraph for agent orchestration with a centralized contextual memory layer. This allows agents to share state and make coordinated decisions based on patient needs.

Frontend

Next.js and TypeScript for the application framework
TailwindCSS and Radix UI for styling and accessible components
Real-time caregiver dashboard for behavioral insights and routine tracking

Impact

Demonstrates how agentic AI systems can deliver practical support in healthcare contexts. The multi-agent approach allows for specialized handling of different care aspects while maintaining a unified patient experience.