PRD.txt

# DermaCheck Educational App - Product Requirements Document

## Executive Summary

**Product Name:** DermaCheck Educational App

**Vision:** An educational mobile application that helps users learn about skin conditions and empowers them to make informed decisions about seeking professional dermatological care.

**Core Principle:** This is an educational tool, NOT a diagnostic device. It teaches users dermatology concepts and helps them track skin changes over time, always directing them to healthcare professionals for actual diagnosis and treatment.

**Competition Entry:** MedGemma Impact Challenge (Kaggle)

**Target Completion:** February 24, 2026

---

## Problem Statement

**Current Challenges:**
- Dermatology wait times average 4-8 weeks in many areas
- People don't know when a skin concern warrants immediate vs. routine consultation
- Lack of skin health education leads to delayed care for serious conditions
- Underserved communities have limited access to dermatology information
- Many people don't know what to look for when self-monitoring moles/lesions

**Our Approach:**
Rather than attempting diagnosis, we educate users about dermatological features and empower them to be better advocates for their skin health when they DO see a professional.

---

## Product Goals & Objectives

### Primary Goals:
1. **Educate** users about common skin conditions and dermatological terminology
2. **Empower** users to track skin changes over time with structured documentation
3. **Guide** users on when to seek professional care (urgency levels)
4. **Reduce** health disparities through accessible skin health education

### Non-Goals (Critical Safety Boundaries):
- ❌ Provide diagnoses
- ❌ Recommend treatments
- ❌ Replace professional medical advice
- ❌ Make clinical decisions
- ❌ Store PHI/medical records (HIPAA complexity)

---

## Target Users

### Primary Personas:

**1. Health-Conscious Individual (25-45)**
- Wants to monitor moles/skin changes
- Proactive about health
- Has smartphone access
- May lack health insurance or dermatology access

**2. Parent/Caregiver (30-50)**
- Monitoring children's rashes, skin conditions
- Wants to know if doctor visit is needed
- Needs educational resources about pediatric skin issues

**3. Underserved Community Member**
- Limited healthcare access
- Language barriers
- May have darker skin tones (often underrepresented in dermatology resources)
- Needs educational materials in multiple languages

---

## Features & Functionality

### MVP Features (Phase 1)

#### 1. **Skin Photo Analysis & Education**
**Functionality:**
- User uploads/takes photo of skin area
- AI describes observable features using dermatological terminology:
  - Symmetry/asymmetry
  - Border regularity
  - Color variation
  - Approximate diameter
  - Surface texture observations
- App explains what each feature means educationally
- Shows similar reference images from educational database
- **Explicitly states: "This is NOT a diagnosis"**

**User Flow:**
```
1. User taps "Learn About a Skin Concern"
2. Takes photo or uploads from gallery
3. Crop/zoom to area of interest
4. AI analyzes observable features
5. Results shown with educational context:
   - "I observe: asymmetrical borders"
   - "What this means: Symmetry is one feature dermatologists examine..."
   - "Similar educational examples: [reference images]"
6. Educational recommendation:
   - "Features observed may warrant professional evaluation"
   - "Schedule appointment with dermatologist"
```

#### 2. **Tracking & Comparison Tool**
**Functionality:**
- Users create "spots" to track over time
- Name/label each tracked spot (e.g., "left shoulder mole")
- Timeline view showing progression
- Side-by-side comparison feature
- Visual change detection highlighting
- Export timeline as PDF for doctor visits

**Safety Features:**
- All stored locally on device (privacy)
- Optional cloud backup (encrypted)
- Data export for medical appointments
- No sharing/social features

#### 3. **Educational Content Library**
**Functionality:**
- Searchable library of skin conditions
- Categories: moles, rashes, infections, inflammatory, cancerous
- Each condition includes:
  - Clinical photos (various skin tones)
  - Description in plain language
  - Common features
  - When to seek care
  - Risk factors
  - **Clear disclaimer this is educational only**

**Content Sections:**
- "Understanding the ABCDE Rule" (melanoma screening)
- "Common Benign Conditions"
- "When to See a Doctor Urgently"
- "Skin Cancer Warning Signs"
- "Skin Types & Sun Protection"

#### 4. **Care Guidance (Triage Education)**
**Functionality:**
Three-tier educational guidance system:

**🔴 Seek Immediate Care:**
- Rapidly growing lesions
- Bleeding that won't stop
- Signs of infection with fever
- Sudden changes in existing moles
- **Action: "Visit emergency room or urgent care today"**

**🟡 Schedule Professional Consultation:**
- New or changing moles
- Persistent rashes
- Unusual skin changes
- Family history of skin cancer
- **Action: "Schedule dermatology appointment within 2-4 weeks"**

**🟢 Monitor & Learn:**
- Stable, long-existing marks
- Minor cosmetic concerns
- General skin health questions
- **Action: "Continue monitoring, educate yourself, mention at next checkup"**

**Critical Safety Feature:**
- System defaults to higher urgency when uncertain
- Always includes: "When in doubt, consult a healthcare professional"

#### 5. **Disclaimers & Safety Messaging**

**On Every Screen:**
- Footer: "Educational tool only - Not for diagnosis"

**Before First Use:**
- Comprehensive disclaimer modal:
  - Purpose: Education and tracking
  - Not a medical device
  - Not a substitute for professional care
  - Accuracy limitations
  - User agreement required

**On Results:**
- Large banner: "These observations are educational only. Consult a dermatologist for diagnosis."
- Link to find dermatologists (via external service)

---

### Phase 2 Features (Post-Competition)

- Multi-language support (Spanish, Mandarin, Hindi)
- Offline mode with downloaded educational content
- Body map for tracking multiple lesions
- Reminders for self-examinations
- Integration with telemedicine platforms
- Accessibility features (voice descriptions)

---

## User Stories

### Core User Journeys:

**Story 1: Learning About a Mole**
```
As a user concerned about a mole,
I want to learn what features dermatologists look for,
So I can make an informed decision about seeking professional care.

Acceptance Criteria:
- User can photograph mole
- App describes observable features with educational context
- App provides ABCDE rule explanation
- App recommends professional consultation for evaluation
- Clear disclaimers present throughout
```

**Story 2: Tracking a Spot Over Time**
```
As a user monitoring a skin spot,
I want to track changes over weeks/months,
So I can share progression data with my dermatologist.

Acceptance Criteria:
- User can create named tracking spots
- Upload multiple photos over time
- View side-by-side comparisons
- See timeline visualization
- Export PDF report for doctor
```

**Story 3: Understanding When to Seek Care**
```
As a parent with a child's rash,
I want to understand urgency levels,
So I know whether to call the doctor today or monitor.

Acceptance Criteria:
- Educational guidance on urgency indicators
- Clear action items for each level
- Examples of conditions requiring immediate vs routine care
- Resources for finding care providers
```

---

## Technical Architecture

### Tech Stack

#### **Frontend - Mobile App**
**Framework:** React Native (cross-platform iOS/Android)
- **Rationale:** Single codebase, good MedGemma integration, large community

**Key Libraries:**
- `react-native-camera` - Photo capture
- `react-native-image-picker` - Gallery access
- `react-native-fs` - Local file storage
- `react-native-vector-icons` - UI icons
- `react-native-pdf` - PDF export
- `@react-navigation` - Navigation

**UI Framework:** React Native Paper (Material Design)

#### **AI/ML Backend**

**Primary Model:** MedGemma 1.5 4B
- **Usage:** Image analysis and feature description
- **Deployment:** Cloud-based API (Vertex AI or HuggingFace Inference)

**Secondary Model:** MedSigLIP (if needed)
- **Usage:** Image encoding/embedding for similarity search

**Inference Options:**
1. **Cloud API** (MVP approach):
   - Google Cloud Vertex AI
   - Lower latency, always up-to-date model
   - Requires internet connection
   - Cost per request

2. **On-Device** (Phase 2):
   - TensorFlow Lite conversion
   - Offline capability
   - Privacy benefits
   - Larger app size, slower inference

**For MVP: Cloud API approach**

#### **Backend Services**

**API Layer:** Node.js + Express
- Handles MedGemma API calls
- Rate limiting
- Usage analytics (anonymous)
- Content delivery

**Database:** Firebase Firestore (optional cloud backup)
- User tracking data (encrypted)
- Sync across devices
- Real-time updates

**Storage:**
- **Primary:** Local device storage (SQLite + filesystem)
- **Optional Backup:** Firebase Storage (encrypted)

**Hosting:**
- Backend API: Google Cloud Run (serverless)
- Static content: Firebase Hosting

#### **Educational Content**

**Content Database:**
- Structured JSON files
- Curated dermatology educational content
- Reference images (open-source medical image datasets)

**Sources:**
- DermNet NZ (open educational resource)
- AAD (American Academy of Dermatology) public materials
- NIH MedlinePlus
- **All properly attributed and licensed**

---

### System Architecture Diagram

```
┌─────────────────────────────────────────┐
│         React Native Mobile App         │
│  ┌─────────────────────────────────┐   │
│  │  UI Components                   │   │
│  │  - Camera Interface              │   │
│  │  - Photo Gallery                 │   │
│  │  - Timeline View                 │   │
│  │  - Educational Content           │   │
│  └─────────────────────────────────┘   │
│               ↓                          │
│  ┌─────────────────────────────────┐   │
│  │  Local Storage (SQLite)          │   │
│  │  - Tracking data                 │   │
│  │  - Photos (encrypted)            │   │
│  │  - User preferences              │   │
│  └─────────────────────────────────┘   │
└─────────────────┬───────────────────────┘
                  │
                  ↓ HTTPS
┌─────────────────────────────────────────┐
│       Backend API (Node.js/Express)     │
│                                          │
│  ┌──────────────────────────────────┐  │
│  │  Endpoints:                       │  │
│  │  - POST /analyze-image            │  │
│  │  - GET /educational-content       │  │
│  │  - POST /generate-report          │  │
│  └──────────────────────────────────┘  │
└────────┬──────────────┬─────────────────┘
         │              │
         ↓              ↓
┌─────────────────┐  ┌──────────────────┐
│  Vertex AI      │  │  Firebase        │
│  (MedGemma)     │  │  - Firestore     │
│  - Image        │  │  - Storage       │
│    Analysis     │  │  - Auth (future) │
│  - Feature      │  │                  │
│    Extraction   │  │                  │
└─────────────────┘  └──────────────────┘
```

---

### MedGemma Integration

#### **Image Analysis Pipeline:**

```python
# Pseudocode for image analysis
def analyze_skin_image(image_path):
    """
    Educational analysis of skin image using MedGemma
    Returns observable features with educational context
    """
    
    # 1. Prepare prompt for MedGemma
    prompt = """
    You are an educational tool helping users learn about skin features.
    Analyze this image and describe OBSERVABLE features only.
    Use dermatological terminology but explain each term.
    
    Describe:
    - Symmetry (is it symmetrical or asymmetrical?)
    - Borders (are they regular, irregular, or poorly defined?)
    - Color (single color or multiple colors? Which?)
    - Diameter (approximate size)
    - Evolution indicators (any signs of change?)
    - Texture (flat, raised, rough, smooth?)
    
    Format response as educational observations, NOT diagnosis.
    End with: "Consult a dermatologist for professional evaluation."
    """
    
    # 2. Call MedGemma API
    response = medgemma_client.analyze(
        image=image_path,
        prompt=prompt,
        max_tokens=500
    )
    
    # 3. Parse response
    features = parse_features(response.text)
    
    # 4. Add educational context
    educational_content = add_context(features)
    
    # 5. Determine guidance level (conservative approach)
    guidance = determine_guidance(features)  # Defaults to "seek consultation"
    
    return {
        "observations": educational_content,
        "guidance": guidance,
        "disclaimer": "This is educational only. Not a diagnosis.",
        "raw_description": response.text
    }
```

#### **API Request Flow:**

1. **User takes photo** → Compress to <2MB
2. **Upload to backend** → Encrypted HTTPS
3. **Backend calls MedGemma** → Vertex AI endpoint
4. **Parse response** → Extract features, add disclaimers
5. **Return to app** → Display with educational context
6. **Save locally** → Store result with photo

#### **Cost Estimation:**
- Vertex AI MedGemma pricing: ~$0.001-0.005 per image
- Target: 10,000 analyses = $10-50 budget
- Optimization: Cache educational content, batch requests

---

## Data & Privacy

### Data Handling

**What We Store Locally:**
- User-captured photos (encrypted)
- Tracking history and labels
- AI analysis results
- User preferences

**What We DON'T Store:**
- Personal identifying information (PII)
- Health records
- Diagnosed conditions
- User names/emails (unless they opt-in for cloud backup)

### Privacy Approach

**Default: Local-First**
- All data stored on device
- No account required
- No cloud upload unless explicitly opted-in

**Optional Cloud Backup:**
- End-to-end encryption
- User controls all data
- Easy data deletion
- Export all data feature

### Compliance Considerations

**Not a Medical Device:**
- Educational tool exemption
- No diagnostic claims
- Clear disclaimers throughout
- FDA registration NOT required (educational use)

**Privacy Laws:**
- GDPR compliant (EU users)
- CCPA compliant (CA users)
- Data minimization approach
- User data control

**Terms of Service:**
- Educational purposes only
- No warranty on accuracy
- User assumes all risk
- Liability limitations

---

## Safety & Risk Management

### Safety Features

#### **1. Conservative Defaults**
- When uncertain → recommend professional consultation
- All unusual features → flag for evaluation
- System bias toward caution

#### **2. Multi-Layer Disclaimers**
- Launch screen agreement
- Every analysis result page
- Export reports
- Settings reminder

#### **3. No False Reassurance**
- Never say "this looks normal"
- Never say "no need to worry"
- Always: "Observable features suggest X, consult professional"

#### **4. Prominent "Get Help" Button**
- Always visible
- Links to:
  - Find a dermatologist (AAD tool)
  - Telehealth options
  - Emergency resources

#### **5. Accuracy Transparency**
- "AI systems can make mistakes"
- "Lighting, angle affect analysis"
- "Always get professional evaluation"

### Risk Mitigation

| Risk | Likelihood | Impact | Mitigation |
|------|------------|--------|------------|
| User treats as diagnosis | High | Severe | Multiple disclaimers, no diagnostic language, always recommend consultation |
| Delayed care from false reassurance | Medium | Severe | Never reassure, always suggest professional evaluation, conservative guidance |
| Poor quality photos | High | Low | Photo quality tips, retake prompts, lighting guidance |
| Misuse for serious conditions | Medium | High | Urgent care guidance prominent, red flag symptoms listed |
| Liability claims | Low | Medium | Strong TOS, educational disclaimers, no medical claims |
| Model bias (skin tone) | Medium | High | Diverse reference images, acknowledge limitations, test across skin types |

---

## Success Metrics

### Competition Metrics

**Primary (for judging):**
- **Impact**: User testimonials, use cases, potential reach
- **Innovation**: Novel use of MedGemma for education
- **Implementation**: Working demo, code quality
- **Safety**: Comprehensive safety approach
- **Presentation**: Clear documentation, demo video

### Product Metrics

**User Engagement:**
- Daily active users (DAU)
- Photos analyzed per user
- Tracking spots created
- Educational content views
- Time spent in app

**Educational Impact:**
- User survey: "Did this help you learn about skin health?"
- "Did this help you decide whether to see a doctor?"
- App store ratings/reviews
- Testimonials

**Safety Metrics:**
- Zero incidents of delayed care (user surveys)
- Disclaimer view rate: 100%
- Professional consultation rate after using app

**Technical:**
- App crashes: <1%
- API response time: <3 seconds
- Photo analysis accuracy: Validated against dermatology textbooks
- Model performance across skin tones: Measure & report

---

## Development Timeline

### Week 1-2: Foundation (Jan 20 - Feb 2)
- [ ] Set up React Native project
- [ ] Design UI/UX mockups
- [ ] Create disclaimer/legal copy
- [ ] Set up Vertex AI MedGemma access
- [ ] Build basic camera interface

### Week 3-4: Core Features (Feb 3 - Feb 16)
- [ ] Implement photo capture/gallery
- [ ] Build MedGemma integration
- [ ] Create analysis results screen with disclaimers
- [ ] Implement local storage (SQLite)
- [ ] Build tracking timeline feature
- [ ] Develop educational content library (10-15 conditions)

### Week 5: Polish & Safety (Feb 17 - Feb 23)
- [ ] Add all disclaimer screens
- [ ] Implement urgency guidance system
- [ ] Create PDF export feature
- [ ] Build onboarding flow
- [ ] User testing with 10-15 people
- [ ] Bug fixes and refinements

### Week 6: Submission (Feb 24)
- [ ] Final testing
- [ ] Create demo video
- [ ] Write documentation
- [ ] Prepare presentation
- [ ] Submit to Kaggle

---

## Competition Deliverables

### Required Submissions:

1. **Working Application**
   - APK file (Android) or TestFlight link (iOS)
   - Source code on GitHub
   - README with setup instructions

2. **Demo Video (3-5 min)**
   - Problem statement
   - Solution walkthrough
   - Key safety features
   - MedGemma integration demo
   - Impact potential

3. **Documentation**
   - Technical architecture
   - MedGemma usage explanation
   - Safety/ethical considerations
   - User testing results
   - Future roadmap

4. **Presentation Deck**
   - Problem & solution
   - Technical approach
   - Demo screenshots
   - Impact metrics
   - Q&A prep

---

## Competitive Advantages

### Why This Could Win:

1. **Safety-First Design**
   - Explicitly educational, not diagnostic
   - Reduces liability while maximizing impact
   - Sets ethical standard

2. **Underserved Focus**
   - Addresses health disparities
   - Multi-skin-tone education
   - Accessible interface

3. **Novel MedGemma Use**
   - Uses multimodal capabilities (images + text)
   - Educational context generation
   - Feature extraction for learning

4. **Practical Utility**
   - Solves real problem (when to see doctor?)
   - Tracking for doctor visits
   - Actionable guidance

5. **Scalability**
   - Mobile-first
   - Minimal backend costs
   - Open-source potential

---

## Future Enhancements (Post-Competition)

### Phase 2 Features:
- Full offline mode with on-device MedGemma
- 10+ language support
- Body map visualization
- Telemedicine integration
- Pediatric skin condition focus
- Accessibility features (screen reader, voice)

### Potential Partnerships:
- Dermatology associations for content validation
- Community health centers for distribution
- Medical schools for education
- Open-source community contribution

### Monetization (if deployed):
- Free base app (educational mission)
- Optional premium: Cloud backup, unlimited tracking, advanced features
- Institutional licenses for clinics/schools
- Grant funding for underserved deployment

---

## Open Questions & Decisions Needed

**Technical:**
- [ ] Which cloud platform? (Google Cloud vs AWS vs Azure)
- [ ] Image size/compression standards?
- [ ] Offline capabilities in MVP?

**Content:**
- [ ] How many skin conditions in initial library?
- [ ] Which languages for Phase 1?
- [ ] Partner with dermatologist for content review?

**Legal:**
- [ ] Need lawyer review of disclaimers?
- [ ] Insurance needed for app?
- [ ] Age restrictions (13+, 18+)?

**UX:**
- [ ] Mandatory tutorial before first use?
- [ ] Gamification elements for engagement?
- [ ] Community features (safe to add)?

---

## Team Roles (if applicable)

**Solo Developer Path:**
- Focus on core MVP features
- Use pre-built UI components
- Leverage MedGemma for intelligence
- Community content for education

**Team Path:**
- **Developer:** React Native app + backend
- **Designer:** UI/UX, educational content design
- **Medical Advisor:** Content validation, safety review
- **Product:** PRD, demo video, presentation

---

## Conclusion

DermaCheck Educational App provides a **safe, impactful, and technically feasible** entry for the MedGemma Impact Challenge. By focusing strictly on education rather than diagnosis, we:

✅ Minimize medical/liability risks
✅ Maximize social impact (health education access)
✅ Showcase MedGemma's multimodal capabilities
✅ Create a scalable, deployable solution
✅ Address real healthcare disparities

The conservative, educational-only approach makes this buildable within the competition timeline while maintaining the highest ethical standards.

---

**Next Steps:**
1. Review and approve this PRD
2. Set up development environment
3. Create detailed UI mockups
4. Begin Week 1 tasks
5. Schedule check-ins (weekly progress reviews)