MemoryBridge

MemoryBridge — AI Memory Companion for Dementia & Early Alzheimer's Care

MemoryBridge is a warm, supportive, fully interactive personal memory companion for people living with mild cognitive impairment, dementia, or early-stage Alzheimer's, and the families who care for them.

It is not a medical, diagnostic, monitoring, or clinical treatment tool. It feels human, warm, and familiar, never like a sterile AI dashboard.

🌟 Core Principle

The AI never invents a person's past.
It speaks strictly and solely from family-approved (verified) memories, offers familiar photos and calming music, and openly states when information is not in the family storybook:
"I don't have that memory in your family storybook yet. Perhaps someone in your family can tell us more."

📐 Architecture & Dual-App Experience

                       ┌──────────────────────────────┐
                       │     MemoryBridge Core DB     │
                       │    (localStorage / Cloud)    │
                       └──────────────┬───────────────┘
                                      │
            ┌─────────────────────────┴─────────────────────────┐
            ▼                                                   ▼
┌───────────────────────────────┐               ┌───────────────────────────────┐
│       Caregiver Web App       │               │      Patient Tablet App       │
│        Routes: /family/*      │               │       Routes: /patient/*      │
├───────────────────────────────┤               ├───────────────────────────────┤
│ • Memory Library & Albums     │               │ • 3D Open-Spread Storybooks   │
│ • 4-Step Story Wizard         │               │ • Big Tactile Face Cards      │
│ • Family Verification Queue   │──── Sync ────►│ • Pressure-Free Puzzles       │
│ • Memory Connection Graph     │   (Verified   │ • Vinyl Record Music Player   │
│ • Timeline & Routine Manager  │    Only)      │ • Bedtime Sleepy Mira Diary   │
│ • Relatives & Bio Directory   │               │ • Conversational Companion    │
└───────────────────────────────┘               └───────────────────────────────┘

🤖 Production AI Conversation Architecture

flowchart TD
    UserQuery["Patient Voice or Text Input"] --> EntityDetection["Topic & Entity Detection\n(People, Places, Years, Events)"]
    EntityDetection --> FilterVerified["Query Database\nFilter: status == 'verified'"]
    
    FilterVerified --> CheckFound{Verified Memory\nFound?}
    
    CheckFound -- "No (Unknown)" --> Fallback["Exact Fallback:\n'I don't have that memory in your family storybook yet.\nPerhaps someone in your family can tell us more.'"]
    
    CheckFound -- "Pending/Unconfirmed" --> PendingGuard["Safety Guard:\n'A family member suggested this,\nbut it has not been confirmed yet.'"]
    
    CheckFound -- "Yes (Verified)" --> PromptBuilder["Build LLM Context\n• Family-Approved Facts Only\n• Zero Extrapolation / No Invented Details\n• Calm, Short Sentences, Reassuring Tone"]
    
    PromptBuilder --> LLM["LLM (e.g. Gemini 1.5 Pro / Flash)"]
    LLM --> AttachMedia["Attach Verified Photographs,\nConnected Song, & Follow-Up Prompts"]
    AttachMedia --> CompanionOutput["Render Companion Speech Bubble\n+ Audio Narration (TTS) + Photo Card"]

🔌 Service Layer & Backend Swap Guide

The application is architected with swap-ready TypeScript interfaces under src/services/index.ts:

1. IMemoryService:

   • getVerifiedMemories(memories): Enforces that patient screens only query memories where status === 'verified'.
   • Swap: Connect to PostgreSQL / Supabase with Row Level Security (RLS) enforcing status = 'verified'.

2. IStoryService:

   • generateStoryDraft(input): Generates deterministic memory drafts from user-entered titles, dates, locations, and tagged people.
   • Swap: Connect to an LLM endpoint with few-shot prompts instructed to never extrapolate beyond passed fields.

3. IConversationService:

   • respond(userText, allMemories, allPeople, patientName): Keyword and entity matching engine with strict safety guardrails.
   • Swap: Connect to a LangChain / FastAPI pipeline running Gemini Flash with strict structured outputs.

4. ISpeechService:

   • Text-to-speech using browser window.speechSynthesis with warm rate (0.88x) and pitch controls.
   • Swap: Connect to Google Cloud Text-to-Speech (Journey or Studio voice models) or ElevenLabs.

5. IVoiceService:

   • Voice speech-to-text using webkitSpeechRecognition with graceful fallback to keyboard input.
   • Swap: Connect to Whisper / Google Cloud Speech-to-Text streaming API.

🎨 Design System & Accessibility

• Primary Colors:
  • Main Brand: #F36B32 (Warm Terracotta Orange)
  • Darker Text-Safe Accent: #B8431A (Passes WCAG AAA contrast against light cream)
  • Backgrounds: #FFFDF9 (Warm White), #F8F1E8 (Warm Cream), #FDEEE4 (Peach Wash), #EBF4F6 (Soft Sky)
• Typography:
  • Headings: Playfair Display & Cormorant Garamond (Editorial, warm serif)
  • Body: Inter & Manrope (Clean, ultra-legible sans-serif)
• Dementia-Friendly UX Rules:
  • 56px+ touch targets on all patient buttons.
  • No countdown timers, no scores, no buzzer sounds, no time pressure.
  • Zero autoplay on audio: music and voice narration start only upon deliberate tap.
  • Text size switcher: Normal (16px+), Large (20px+), Extra Large (24px+).
  • Reduced motion support via prefers-reduced-motion in CSS.

🚀 Running the Prototype Locally

# 1. Install dependencies
npm install

# 2. Run development server
npm run dev

# 3. Open in browser
http://127.0.0.1:5173/

• Navigate to / for the Public Landing Page.
• Navigate to /patient for the Tablet Companion Experience.
• Navigate to /family for the Caregiver Management Hub.
• Edits, new storybooks, and verification approvals persist in localStorage.
• Click "Reset Demo Data" in Caregiver Settings anytime to restore original state.