EverMind

Banner
Story 1
Story 2
Vision
Architecture

Inspiration

EverMind was inspired by our personal experience caring for a family member with Alzheimer's disease. Through this journey, we witnessed the emotional, mental, and daily challenges faced by both patients and caregivers. To better understand these struggles, we explored research studies and articles on dementia care and digital health, which helped us identify gaps where technology could offer meaningful support.

Research has shown that digital health interventions can significantly reduce caregiver burden and improve quality of life for both patients and caregivers [1]. We found that while many apps focus on medication reminders and tracking, few address the immediate emotional crisis moments that patients experience, including episodes of confusion, anxiety, or disorientation that require instant, personalized intervention.

What it does

EverMind is designed to support both patients and caregivers by creating a connected, compassionate care environment. The app allows caregivers and patients to stay connected while simplifying daily care tasks. Key features include:

Patient Profile Management: Connecting caregivers and patients in one platform with detailed profiles including core identity, comfort memories, safe places, and behavioral patterns
Activity Tracking: Setting up reminders for daily tasks such as taking medication, drinking water, and routine activities
Mood & Behavior Monitoring: Tracking the patient's condition and disease stage through a calendar-based system with mood logs and caregiver notes
Emergency SOS Voice Assistant: AI-powered conversational support that provides immediate grounding techniques during crisis moments using real-time patient context
AI Chat Companion: Offering emotional support, companionship, and gentle mental engagement for patients through voice-based conversations

Our emergency grounding feature is particularly innovative, when a patient experiences confusion or distress, caregivers can trigger an instant voice conversation with an AI assistant that has complete context about the patient's background, recent mood patterns, comfort topics, and proven calming strategies.

How we built it

EverMind is built on a sophisticated three-tier architecture that prioritizes patient safety and personalized care:

Frontend Layer

React 18 with React Router v6 for a responsive, single-page application
React Context API for global patient state management
localStorage for client-side data persistence
Tailwind CSS for accessible, dementia-friendly design following best practices [2]

Data Processing Layer - The MCP Innovation

The core innovation of EverMind is our custom MCP (Model Context Protocol) Server that acts as an intelligence hub:

Activity Tracker: Stores and retrieves mood logs, caregiver notes, and behavioral patterns
Profile Mapper: The mapPatientToProfile() function combines static patient profiles with dynamic tracking data
Context Prioritization: Emergency notes are automatically surfaced first, ensuring the AI responds appropriately to critical situations
Prompt Generator: Builds comprehensive system prompts with evidence-based grounding techniques validated by research [1]

Scientific research guided our implementation of five key grounding techniques:

Validation & Reassurance - Acknowledging feelings without correction
Reminiscence Therapy - Using personal memories for comfort
Sensory Grounding - Engaging present-moment awareness
Redirection - Gentle topic changes to calming subjects
Environmental Anchoring - Reinforcing safety and familiarity

These techniques are embedded directly into our AI prompts based on peer-reviewed studies on dementia care interventions [1].

AI Voice Layer

VAPI Platform: Orchestrates real-time voice conversations
Claude Sonnet 4 (claude-sonnet-4-20250514): Anthropic's AI model powers context-aware, empathetic responses
Deepgram Nova 2: Provides real-time speech-to-text transcription
Azure Neural Voices (Jenny/Guy): Delivers natural, warm text-to-speech output

Data Flow Architecture

The emergency SOS flow demonstrates our MCP-driven approach:

Caregiver triggers SOS → EverMind Client retrieves PatientContext + recent tracking data
Data sent via REST API to EverMind MCP Server
MCP Server maps data to structured profile using MCP Protocol
Profile + system prompt sent to VAPI via HTTPS
Real-time voice pipeline: Patient speaks → Deepgram (WebSocket) → Claude (API) → Azure TTS (API)
Audio + transcript returned via WebSocket to patient

This architecture ensures that every AI response is grounded in the patient's unique history, current emotional state, and proven calming strategies. All will be delivered within seconds of an emergency trigger.

Design Considerations

Following dementia-friendly design principles [2, 3], we implemented:

Large, clear text (minimum 16px)
High contrast color schemes
Simple, consistent navigation
Minimal cognitive load with focused interfaces
Reassuring visual language

Challenges we ran into

One of our biggest challenges was identifying features that genuinely help patients rather than simply adding complexity. We constantly questioned whether each feature would truly improve a patient's daily life. Research shows that apps for dementia patients must balance functionality with simplicity [3], which guided our decision to focus on core features rather than feature bloat.

Technical challenges included:

Learning and implementing the MCP Protocol for the first time
Designing a prompt structure that balances empathy with safety guidelines
Managing real-time WebSocket connections for voice streaming
Ensuring VAPI configuration properly included all patient context
Testing grounding techniques without access to real patients (we used research-validated methods instead [1])

Additionally, this was our first time building and filming a project like this. Learning new tools like VAPI, working with voice AI APIs, and understanding the nuances of conversational design pushed us to grow significantly as a team.

Accomplishments that we're proud of

We are especially proud of the emergency grounding feature, which provides an added layer of safety for patients and peace of mind for caregivers. The MCP Server architecture ensures that the AI has complete, real-time context, including mood trends from the past week and today's caregiver notes, enabling truly personalized crisis support.

Another major accomplishment is the AI chat companion, which focuses on supporting patients' mental and emotional well-being by offering companionship and gentle interaction. Unlike generic chatbots, our system is trained on patient-specific calming topics and designed to avoid triggering subjects.

Technical achievements:

Successfully implemented a custom MCP Server that structures patient data for optimal AI consumption
Created a seamless voice pipeline integrating three different AI services (Deepgram, Claude, Azure)
Built a tracking system that prioritizes emergency information using tag-based categorization ([emergency], [behavior], [medical])
Designed prompts that generate empathetic, grounding responses validated against clinical research

What we learned

Through this project, we learned that truly helping patients means seeing the world from their perspective. Empathy, patience, and simplicity are just as important as technical innovation when designing healthcare solutions.

Key technical learnings:

MCP Protocol is powerful for providing structured context to AI models
Voice AI requires careful orchestration of multiple services with proper error handling
Real-time systems need robust WebSocket management and event handling
Prompt engineering for healthcare requires balancing warmth with safety

Key design learnings:

Every feature must answer: "Does this reduce confusion or add it?"
Research-backed techniques (like reminiscence therapy) are more effective than intuition
Dementia-friendly design principles significantly impact usability [2]

What's next for EverMind

Regardless of the competition outcome, we plan to continue improving EverMind. Our roadmap includes:

Short-term:

User testing with real caregivers and patients (with proper consent and oversight)
Adding more language support for diverse communities
Implementing medication reminders with voice confirmations
Expanding the tracking system to include sleep patterns and physical activities

Long-term:

Integration with wearable devices for passive health monitoring
Caregiver support groups and resource sharing within the app
Collaboration with healthcare providers for clinical validation
Research partnership to measure impact on caregiver burden and patient quality of life

We will keep learning, refining features, and listening to real user needs to make the app more supportive, accessible, and impactful for families affected by Alzheimer's disease.

References

[1] Boots, L. M., de Vugt, M. E., van Knippenberg, R. J., Kempen, G. I., & Verhey, F. R. (2014). A systematic review of Internet-based supportive interventions for caregivers of patients with dementia. International Journal of Geriatric Psychiatry, 29(4), 331–344. https://doi.org/10.1002/gps.4016

[2] Alzheimer's Society. Why is digital design important for someone affected by dementia? https://www.alzheimers.org.uk/blog/how-design-website-someone-affected-dementia

[3] Topflight Apps. Inside the Mind: What It Takes to Build a Truly Helpful Dementia App. https://topflightapps.com/ideas/alzheimers-dementia-prevention-app-development/