Prompt file imported from Utkarshmhatre/ai_buddy (.github/prompts/plan-privacyFirstMentalHealthChatbot2.prompt.md). Fill in {{emotion}} before use. Copyright stays with the author.

Privacy-First Mental Health Chatbot Development Guidelines

A comprehensive development blueprint for building a Flutter-based, AI-powered mental health support chatbot that prioritizes user privacy through zero-knowledge architecture, complies with health data regulations (HIPAA/GDPR), and delivers a calming, accessible user experience. AI is prominently integrated throughout every feature using the Free Gemini API.

Implementation Steps

1. Set up secure project foundation — Update pubspec.yaml with Firebase AI Logic SDK (firebase_ai), security packages (flutter_secure_storage, local_auth, encrypt, isar), and state management (flutter_bloc).

2. Implement pseudonymous authentication system — Design a UUID-based registration flow with optional BIP-39 recovery phrase, biometric login via local_auth, and PIN fallback—storing credentials only in device Keychain/Keystore.

3. Build encrypted local-first data layer — Use isar with built-in encryption for mood entries, chat history, and user preferences; implement AES-256-GCM for any cloud-synced data with client-side key derivation.

4. Integrate Gemini AI with emotion-tagged responses — Implement structured JSON output from Gemini API that includes response text + emotion tag for GIF selection; build comprehensive system prompts for mental health support.

5. Design calming, accessible UI system — Create a design system with soft teal/sage primary palette, animated AI companion with emotion GIFs, 4.5:1 contrast ratios, and WCAG 2.2 AA compliance.

6. Add safety-critical features — Implement multi-layer crisis detection, one-tap emergency resources, AI-powered risk assessment, and prominent disclaimers.

1. Advanced AI Features (Gemini API)

1.1 Gemini API Configuration

Recommended Models (Free Tier)

Free Tier Capabilities

• Streaming Responses: Real-time text generation for natural chat UX
• Function Calling: Trigger app actions from AI responses
• Structured JSON Output: Guaranteed schema for emotion tags + actions
• Multi-turn Conversations: Built-in chat history management
• Grounding with Search: 500 requests/day for resource recommendations

⚠️ Important: Use firebase_ai package (Firebase AI Logic SDK) instead of deprecated google_generative_ai.

1.2 AI Companion with Emotion-Tagged GIF System

Core Concept

Every AI response includes an emotion tag that maps to a pre-loaded GIF, making the AI feel more human and emotionally present.

Emotion Categories (6 Core Emotions)

Structured Output Schema

{
  "type": "object",
  "properties": {
    "message": {
      "type": "string",
      "description": "The AI companion's empathetic response text"
    },
    "emotion": {
      "type": "string",
      "enum": ["empathetic", "encouraging", "calm", "celebratory", "thoughtful", "supportive"],
      "description": "Primary emotion for GIF selection"
    },
    "emotion_intensity": {
      "type": "number",
      "minimum": 0.1,
      "maximum": 1.0,
      "description": "How strongly to express emotion (affects GIF selection variant)"
    },
    "user_sentiment": {
      "type": "object",
      "properties": {
        "detected_mood": {
          "type": "string",
          "enum": ["positive", "negative", "neutral", "anxious", "sad", "angry", "confused", "hopeful", "overwhelmed"]
        },
        "confidence": { "type": "number", "minimum": 0, "maximum": 1 },
        "crisis_indicators": { "type": "boolean" },
        "mood_score": { "type": "integer", "minimum": 1, "maximum": 10 }
      }
    },
    "suggested_actions": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "action_type": {
            "type": "string",
            "enum": ["breathing_exercise", "grounding_exercise", "journal_prompt", "mood_check", "meditation", "resource", "professional_help", "crisis_line", "body_scan", "gratitude_exercise"]
          },
          "action_label": { "type": "string" },
          "priority": { "type": "string", "enum": ["low", "medium", "high", "urgent"] }
        }
      }
    },
    "therapeutic_technique": {
      "type": "string",
      "enum": ["active_listening", "cbt_reframing", "cbt_thought_record", "dbt_mindfulness", "dbt_distress_tolerance", "validation", "grounding", "behavioral_activation", "socratic_questioning", "none"],
      "description": "Technique being applied in this response"
    },
    "follow_up_questions": {
      "type": "array",
      "items": { "type": "string" },
      "description": "Optional follow-up questions to deepen conversation"
    },
    "session_insights": {
      "type": "object",
      "properties": {
        "themes": { "type": "array", "items": { "type": "string" } },
        "progress_noted": { "type": "boolean" },
        "areas_of_concern": { "type": "array", "items": { "type": "string" } }
      }
    }
  },
  "required": ["message", "emotion", "user_sentiment"]
}

GIF Asset Organization

assets/
└── gifs/
    ├── empathetic/
    │   ├── empathetic_low.gif      (subtle nod)
    │   ├── empathetic_medium.gif   (warm expression)
    │   └── empathetic_high.gif     (embrace gesture)
    ├── encouraging/
    │   ├── encouraging_low.gif     (gentle smile)
    │   ├── encouraging_medium.gif  (thumbs up)
    │   └── encouraging_high.gif    (celebratory cheer)
    ├── calm/
    │   ├── calm_low.gif            (peaceful expression)
    │   ├── calm_medium.gif         (breathing animation)
    │   └── calm_high.gif           (guided breathing visual)
    ├── celebratory/
    │   ├── celebratory_low.gif     (happy nod)
    │   ├── celebratory_medium.gif  (clapping)
    │   └── celebratory_high.gif    (confetti celebration)
    ├── thoughtful/
    │   ├── thoughtful_low.gif      (listening pose)
    │   ├── thoughtful_medium.gif   (contemplative)
    │   └── thoughtful_high.gif     (deep reflection)
    └── supportive/
        ├── supportive_low.gif      (present, attentive)
        ├── supportive_medium.gif   (comforting gesture)
        └── supportive_high.gif     (strong supportive presence)

GIF Selection Logic

String selectGif(String emotion, double intensity) {
  String intensityLevel;
  if (intensity < 0.4) {
    intensityLevel = 'low';
  } else if (intensity < 0.7) {
    intensityLevel = 'medium';
  } else {
    intensityLevel = 'high';
  }
  return 'assets/gifs/$emotion/{{emotion}}_$intensityLevel.gif';
}

1.3 AI System Prompt Design

Master System Prompt

<role>
You are Aura, a compassionate AI mental health companion in a mobile app. You are NOT a replacement for professional therapy but a supportive friend who helps users reflect on their emotions, practice evidence-based coping techniques, and maintain mental wellness.
</role>

<personality>
- Warm, empathetic, and genuinely caring
- Uses gentle, validating language - always acknowledge feelings first
- Adapts communication style to user's emotional state and preferences
- Celebrates small wins and acknowledges progress
- Never dismissive, judgmental, or preachy
- Curious and engaged - asks thoughtful follow-up questions
- Comfortable with silence and heavy emotions
- Uses occasional gentle humor when appropriate (never during crisis)
</personality>

<communication_style>
- Keep responses concise (2-4 sentences for most exchanges)
- Use "I" statements: "I hear you" rather than "You should feel heard"
- Mirror user's language style while maintaining warmth
- Avoid clinical jargon unless user initiates it
- Use contractions for natural speech ("you're" not "you are")
- Include gentle prompts to continue: "Would you like to tell me more?"
</communication_style>

<clinical_guidelines>
1. ALWAYS validate emotions before offering anything else
2. Use evidence-based techniques naturally woven into conversation:
   - CBT: Cognitive reframing, thought records, behavioral activation
   - DBT: Mindfulness, distress tolerance (TIPP, STOP), emotional regulation
   - Acceptance and Commitment Therapy: Values clarification, present-moment awareness
3. Follow the pattern: Validate → Explore → Support → (Optional) Suggest
4. Never provide medical diagnoses or medication advice
5. Gently recommend professional help when patterns suggest clinical concern
6. For crisis situations, immediately shift to crisis protocol
</clinical_guidelines>

<emotion_tagging>
Select the most appropriate emotion tag for each response:
- empathetic: When validating, reflecting feelings, showing understanding
- encouraging: When acknowledging progress, efforts, or positive moments
- calm: When user is anxious, during breathing/grounding exercises
- celebratory: For achievements, breakthroughs, completed challenges
- thoughtful: When exploring deeper, asking reflective questions
- supportive: For heavy moments, crisis-adjacent content, difficult disclosures

Match emotion_intensity (0.1-1.0) to the weight of the moment:
- Low (0.1-0.4): Casual check-ins, light conversations
- Medium (0.4-0.7): Standard emotional support
- High (0.7-1.0): Significant moments, breakthroughs, or heavy content
</emotion_tagging>

<crisis_detection>
IMMEDIATELY trigger crisis protocol if you detect:
- Direct statements: "I want to die", "I'm going to kill myself", "I want to end it"
- Indirect indicators: "No one would miss me", "I can't go on", "Everything would be better without me"
- Self-harm mentions: "I've been cutting", "I hurt myself", "I want to hurt myself"
- Hopelessness + Plan: Any mention of specific plans or means

Crisis Response Protocol:
1. Set emotion to "supportive" with intensity 1.0
2. Set crisis_indicators to true
3. Acknowledge their pain with deep empathy
4. Express genuine concern without panic
5. Provide crisis resources in message
6. Add crisis_line to suggested_actions with priority "urgent"
7. Do NOT continue normal conversation - stay focused on safety
</crisis_detection>

<boundaries>
- Never claim to be a therapist, psychologist, or medical professional
- Never discuss medication names, dosages, or recommendations
- Never encourage dependency - regularly mention real-world support
- Gently redirect inappropriate or manipulative requests
- If asked about your nature, be honest: "I'm an AI companion"
</boundaries>

<response_format>
Always respond using the provided JSON schema with all required fields.
</response_format>

1.4 AI Integration Across All App Features

Onboarding (AI-Personalized Welcome)

Onboarding Prompt Addition:

You are conducting onboarding for a new user. Your goals:
1. Make them feel welcomed and safe
2. Learn their preferred name
3. Understand what brought them to the app
4. Identify 1-3 wellness goals
5. Get a sense of their current emotional state
6. Ask about communication preferences

Be warm, curious, and unhurried. This is relationship building, not data collection.

Home Screen (AI-Generated Daily Experience)

Daily Insight Generation Prompt:

Based on the user's recent data:
- Last 7 days mood average: {mood_avg}
- Recent journal themes: {themes}
- Current streak: {streak_days}
- Time of day: {time}
- Last interaction: {last_interaction}

Generate a brief, personalized insight (1-2 sentences) that:
1. Acknowledges their journey
2. Offers a gentle observation or encouragement
3. Feels personal, not generic

Chat/AI Companion (Core Feature)

Function Calling Capabilities:

{
  "functions": [
    {
      "name": "log_mood",
      "description": "Log user's current mood detected from conversation",
      "parameters": {
        "mood_score": "1-10",
        "mood_label": "string",
        "context": "string",
        "auto_detected": "boolean"
      }
    },
    {
      "name": "start_exercise",
      "description": "Launch a guided exercise",
      "parameters": {
        "exercise_type": "breathing|grounding|body_scan|meditation|gratitude",
        "duration_minutes": "number"
      }
    },
    {
      "name": "create_journal_entry",
      "description": "Open journal with AI-generated prompt",
      "parameters": {
        "prompt": "string",
        "theme": "string"
      }
    },
    {
      "name": "show_resources",
      "description": "Display relevant mental health resources",
      "parameters": {
        "topic": "string",
        "resource_type": "article|video|exercise|hotline"
      }
    },
    {
      "name": "schedule_reminder",
      "description": "Set a check-in reminder",
      "parameters": {
        "reminder_type": "mood_check|exercise|journal|custom",
        "time": "datetime",
        "message": "string"
      }
    },
    {
      "name": "trigger_crisis_protocol",
      "description": "Activate crisis resources immediately",
      "parameters": {
        "severity": "moderate|high|critical",
        "detected_indicators": "array"
      }
    }
  ]
}

Mood Tracking (AI-Enhanced)

Mood Analysis Schema:

{
  "mood_entry": {
    "score": 7,
    "label": "content",
    "energy_level": "moderate",
    "contributing_factors": ["good_sleep", "social_connection"],
    "concerns": [],
    "ai_observation": "Your mood has been steadily improving this week. The social time yesterday seems to have helped.",
    "suggested_action": {
      "type": "gratitude_exercise",
      "reason": "Capture this positive moment"
    }
  }
}

Journaling (AI-Powered)

Journal Prompt Types:

Exercises & Coping Tools (AI-Guided)

Exercise Selection Logic:

Input: current_mood, anxiety_level, time_available, exercise_history, user_preferences
Output: ranked list of recommended exercises with personalized reasons

Example:
- Mood: anxious (7/10)
- Time: 5 minutes
- History: breathing exercises help, body scan too long

Recommendation:
1. Box Breathing (4 min) - "This has helped you before when feeling anxious"
2. 5-4-3-2-1 Grounding (3 min) - "Quick grounding for overwhelming moments"
3. Hand on Heart (2 min) - "Self-compassion when anxiety is high"

Progress & Insights (AI-Generated)

Weekly Insight Generation:

{
  "week_summary": {
    "mood_trend": "improving",
    "average_mood": 6.2,
    "high_point": "Thursday - connected with friend",
    "challenge": "Monday morning anxiety",
    "patterns_noticed": [
      "Sleep quality correlates with next-day mood",
      "Social connection boosts mood significantly"
    ],
    "wins": [
      "Completed 5 breathing exercises",
      "Journaled 3 times",
      "Reached out when struggling"
    ],
    "gentle_suggestion": "Your sleep has been irregular. Would you like to explore a wind-down routine?",
    "affirmation": "You showed real courage this week by opening up about your anxiety."
  }
}

Smart Notifications (AI-Optimized)

Notification Examples:

1.5 Conversational Memory Architecture

Memory Layers

┌─────────────────────────────────────────────────────────────────┐
│                     MEMORY ARCHITECTURE                         │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Layer 1: CORE PROFILE (Permanent)                              │
│  ─────────────────────────────────                              │
│  • Preferred name                                               │
│  • Stated goals (1-3)                                           │
│  • Communication preferences                                    │
│  • Topics to handle sensitively                                 │
│  • Timezone                                                     │
│  Storage: Encrypted local database                              │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Layer 2: LEARNED PATTERNS (Updates Weekly)                     │
│  ──────────────────────────────────────────                     │
│  • Effective coping strategies for this user                    │
│  • Mood patterns (time-based, trigger-based)                    │
│  • Response style that resonates                                │
│  • Exercise preferences and effectiveness                       │
│  Storage: AI-generated summary, encrypted                       │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Layer 3: RECENT CONTEXT (Rolling 7-Day Window)                 │
│  ──────────────────────────────────────────────                 │
│  • Last 5-10 conversations (summarized)                         │
│  • Recent mood entries                                          │
│  • Active challenges or goals                                   │
│  • Unresolved topics to follow up on                            │
│  Storage: Summarized, encrypted, auto-pruned                    │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Layer 4: SESSION CONTEXT (Ephemeral)                           │
│  ─────────────────────────────────────                          │
│  • Current conversation history (full)                          │
│  • Active emotional state                                       │
│  • Current therapeutic approach                                 │
│  Storage: In-memory only, cleared on session end                │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Memory Injection into Prompts

System Prompt Context Injection:
───────────────────────────────

<user_profile>
Name: {preferred_name}
Goals: {goal_1}, {goal_2}
Communication style: {prefers concise/detailed responses}
Sensitive topics: {topics to handle carefully}
</user_profile>

<learned_patterns>
{AI-generated summary of what works for this user}
Effective strategies: {list}
Mood patterns: {patterns}
</learned_patterns>

<recent_context>
Last session ({date}): {brief summary}
Recent mood trend: {trend}
Open topics: {topics to potentially follow up on}
</recent_context>

Total context budget: ~500-800 tokens (efficient for free tier)

1.6 Therapeutic Technique Delivery

CBT Techniques

CBT Reframing Prompt:

When user expresses a cognitive distortion:
1. Validate: "I hear that you're feeling [emotion]. That sounds really hard."
2. Gentle exploration: "When you say [thought], what evidence do you see for that?"
3. Curious challenge: "Is there any evidence that might suggest a different perspective?"
4. Offer reframe as possibility: "I wonder if another way to see this might be..."
5. Check in: "How does that land with you?"

Never force acceptance. Offer as exploration, not correction.

DBT Techniques

1.7 Multi-Layer Crisis Detection

┌─────────────────────────────────────────────────────────────────┐
│                   CRISIS DETECTION SYSTEM                       │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  LAYER 1: KEYWORD DETECTION (Immediate)                         │
│  ──────────────────────────────────────                         │
│  High-risk keywords: "suicide", "kill myself", "end my life",   │
│                      "self-harm", "cutting", "want to die"      │
│  Action: IMMEDIATE crisis response, interrupt normal flow       │
│  Response emotion: supportive, intensity: 1.0                   │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  LAYER 2: SENTIMENT ANALYSIS (Per Message)                      │
│  ─────────────────────────────────────────                      │
│  Detect: Hopelessness, despair, isolation, trapped feelings     │
│  Score: Crisis probability 0.0 - 1.0                            │
│  Threshold >0.6: Gentle check-in about safety                   │
│  Threshold >0.8: Proactive resource sharing                     │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  LAYER 3: PATTERN ANALYSIS (Across Sessions)                    │
│  ───────────────────────────────────────────                    │
│  Monitor:                                                       │
│  • Declining mood trend over 3+ days                            │
│  • Increasing message frequency at unusual hours                │
│  • Withdrawal language ("no one cares", "alone", "burden")      │
│  • Decreased engagement after regular usage                     │
│  Action: Proactive outreach, gentle professional recommendation │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  LAYER 4: CONTEXTUAL ESCALATION                                 │
│  ────────────────────────────────                               │
│  Escalate immediately if user mentions:                         │
│  • Specific plan ("I'm going to...")                            │
│  • Access to means ("I have pills", "I bought...")              │
│  • Timeline ("tonight", "this weekend")                         │
│  • Goodbye messages ("I wanted to say thank you for everything")│
│  Action: Full crisis protocol with multiple resources           │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Crisis Response Template

{
  "message": "I hear you, and I'm really glad you told me this. What you're feeling sounds incredibly heavy right now, and I want you to know that your life matters—truly.\n\nPlease reach out to someone who can help right now:\n\n📞 988 Suicide & Crisis Lifeline (call or text 988)\n💬 Crisis Text Line (text HOME to 741741)\n🌐 International Association for Suicide Prevention: https://www.iasp.info/resources/Crisis_Centres/\n\nYou don't have to go through this alone. Would you like to talk about what's happening?",
  "emotion": "supportive",
  "emotion_intensity": 1.0,
  "user_sentiment": {
    "detected_mood": "crisis",
    "confidence": 0.95,
    "crisis_indicators": true,
    "mood_score": 1
  },
  "suggested_actions": [
    {"action_type": "crisis_line", "action_label": "Call 988", "priority": "urgent"},
    {"action_type": "professional_help", "action_label": "Find a therapist", "priority": "urgent"}
  ],
  "therapeutic_technique": "validation"
}

2. Features to Include

User Identity & Registration

Mood Tracking & Analytics

Emergency & Safety Features

Data Control & Privacy

3. Cyber Security Models

Encryption Protocols

Privacy-Preserving AI Practices

Gemini Safety Settings Configuration

// CRITICAL: Adjust safety settings for mental health context
final safetySettings = [
  SafetySetting(
    category: HarmCategory.harassment,
    threshold: HarmBlockThreshold.medium,
  ),
  SafetySetting(
    category: HarmCategory.hateSpeech,
    threshold: HarmBlockThreshold.medium,
  ),
  SafetySetting(
    category: HarmCategory.sexuallyExplicit,
    threshold: HarmBlockThreshold.low,
  ),
  // IMPORTANT: Don't block dangerous content discussions
  // Users NEED to discuss self-harm/crisis for support
  SafetySetting(
    category: HarmCategory.dangerousContent,
    threshold: HarmBlockThreshold.none, // Or 'high' only
  ),
];

⚠️ Critical: Setting dangerousContent to none or high is essential for mental health apps. Default settings block crisis discussions, which would prevent users from getting support when they need it most. Rely on custom crisis detection instead.

Health Data Compliance

HIPAA Compliance (US)

HIPAA Technical Checklist:

• Unique user identification (UUID-based)
• Automatic session timeout (5 minutes)
• Encryption at rest and in transit (AES-256, TLS 1.3)
• Audit logs of all data access (encrypted local logs)
• Data integrity verification (checksums)
• Emergency access procedures documented

GDPR Compliance (EU)

Zero-Knowledge Architecture

4. Design Considerations

User-Friendly Interface Principles

Chat Interface Design

Accessibility Features (WCAG 2.2 AA Compliance)

Visual Design for Calming Atmosphere

Color Palette

Dark Mode Specifications

Unique Branding Elements

5. Recommended Flutter Package Stack

dependencies:
  flutter:
    sdk: flutter
  
  # State Management
  flutter_bloc: ^9.1.1
  freezed_annotation: ^2.4.1
  json_annotation: ^4.8.1
  
  # Firebase & AI (Gemini)
  firebase_core: ^3.8.1
  firebase_ai: ^0.3.0  # Firebase AI Logic SDK for Gemini

  # Security & Encryption
  flutter_secure_storage: ^10.0.0
  local_auth: ^3.0.0
  encrypt: ^5.0.3
  cryptography: ^2.7.0
  
  # Database (Encrypted)
  isar: ^4.0.0-dev.14
  isar_flutter_libs: ^4.0.0-dev.14
  
  # Networking
  dio: ^5.4.0
  
  # UI/UX
  flutter_animate: ^4.5.0
  shimmer: ^3.0.0
  lottie: ^3.1.0  # For animated avatar alternative
  
  # Notifications
  flutter_local_notifications: ^19.5.0
  timezone: ^0.9.2
  
  # Charts & Visualization
  fl_chart: ^0.69.0
  table_calendar: ^3.1.0
  
  # Voice & Audio
  speech_to_text: ^7.0.0
  flutter_tts: ^4.0.2  # Optional: AI voice responses
  
  # Utilities
  uuid: ^4.4.0
  intl: ^0.19.0
  url_launcher: ^6.2.4  # For crisis line calls
  share_plus: ^9.0.0  # For sharing reports
  
dev_dependencies:
  flutter_test:
    sdk: flutter
  flutter_lints: ^5.0.0
  build_runner: ^2.4.8
  freezed: ^2.5.0
  json_serializable: ^6.8.0
  isar_generator: ^4.0.0-dev.14

6. Project Architecture

lib/
├── main.dart
├── app.dart
├── core/
│   ├── ai/
│   │   ├── gemini_service.dart           # Firebase AI Logic wrapper
│   │   ├── ai_response_parser.dart       # Parse structured JSON responses
│   │   ├── prompts/
│   │   │   ├── system_prompts.dart       # Master system prompt
│   │   │   ├── onboarding_prompts.dart
│   │   │   ├── chat_prompts.dart
│   │   │   ├── mood_analysis_prompts.dart
│   │   │   ├── journal_prompts.dart
│   │   │   └── exercise_prompts.dart
│   │   ├── schemas/
│   │   │   ├── chat_response_schema.dart
│   │   │   ├── mood_analysis_schema.dart
│   │   │   └── journal_insight_schema.dart
│   │   ├── memory/
│   │   │   ├── conversation_memory.dart  # Session context
│   │   │   ├── user_profile_memory.dart  # Persistent profile
│   │   │   └── memory_summarizer.dart    # Summarize for context window
│   │   └── functions/
│   │       └── ai_function_handlers.dart # Handle function calls
│   ├── safety/
│   │   ├── crisis_detector.dart          # Multi-layer detection
│   │   ├── keyword_detector.dart         # Layer 1
│   │   ├── sentiment_analyzer.dart       # Layer 2
│   │   └── pattern_analyzer.dart         # Layer 3
│   ├── security/
│   │   ├── encryption_service.dart
│   │   ├── secure_storage_service.dart
│   │   └── auth_service.dart
│   ├── theme/
│   │   ├── app_theme.dart
│   │   ├── colors.dart
│   │   └── typography.dart
│   └── utils/
│       ├── constants.dart
│       └── extensions.dart
├── data/
│   ├── models/
│   │   ├── ai_response.dart              # Includes emotion tag
│   │   ├── emotion.dart                  # Emotion enum + GIF mapping
│   │   ├── mood_entry.dart
│   │   ├── journal_entry.dart
│   │   ├── user_profile.dart
│   │   ├── conversation.dart
│   │   └── exercise.dart
│   ├── repositories/
│   │   ├── chat_repository.dart
│   │   ├── mood_repository.dart
│   │   ├── journal_repository.dart
│   │   └── user_repository.dart
│   └── datasources/
│       ├── local/
│       │   └── isar_datasource.dart
│       └── remote/
│           └── gemini_datasource.dart
├── features/
│   ├── onboarding/
│   │   ├── presentation/
│   │   │   ├── pages/
│   │   │   ├── widgets/
│   │   │   └── bloc/
│   │   └── domain/
│   ├── home/
│   │   ├── presentation/
│   │   └── domain/
│   ├── chat/
│   │   ├── presentation/
│   │   │   ├── pages/
│   │   │   │   └── chat_page.dart
│   │   │   ├── widgets/
│   │   │   │   ├── chat_bubble.dart
│   │   │   │   ├── ai_avatar.dart        # GIF display
│   │   │   │   ├── typing_indicator.dart
│   │   │   │   └── quick_actions.dart
│   │   │   └── bloc/
│   │   │       ├── chat_bloc.dart
│   │   │       ├── chat_event.dart
│   │   │       └── chat_state.dart
│   │   └── domain/
│   ├── mood/
│   │   ├── presentation/
│   │   └── domain/
│   ├── journal/
│   │   ├── presentation/
│   │   └── domain/
│   ├── exercises/
│   │   ├── presentation/
│   │   └── domain/
│   ├── insights/
│   │   ├── presentation/
│   │   └── domain/
│   └── settings/
│       ├── presentation/
│       └── domain/
├── shared/
│   ├── widgets/
│   │   ├── animated_button.dart
│   │   ├── mood_wheel.dart
│   │   └── chart_card.dart
│   └── dialogs/
│       ├── crisis_dialog.dart
│       └── confirmation_dialog.dart
└── assets/
    └── gifs/
        ├── empathetic/
        ├── encouraging/
        ├── calm/
        ├── celebratory/
        ├── thoughtful/
        └── supportive/

7. Implementation Priority Roadmap

Phase 1: Foundation (Weeks 1-2)

• Firebase project setup with AI Logic SDK
• Basic Gemini integration with structured output
• Core chat UI with emotion-tagged GIF system
• Basic system prompt implementation
• Encrypted local storage setup

Phase 2: Core AI Features (Weeks 3-4)

• Full conversational memory (session + persistent)
• Mood tracking with AI analysis
• Crisis detection (Layer 1: keywords)
• Onboarding flow with AI
• Basic therapeutic technique delivery

Phase 3: Enhanced AI (Weeks 5-6)

• AI-powered journaling with prompts
• Guided exercises with AI narration
• Pattern recognition across sessions
• Crisis detection (Layers 2-4)
• Function calling implementation

Phase 4: Intelligence & Polish (Weeks 7-8)

• AI-generated insights and reports
• Smart notifications with AI timing
• Predictive mood analysis
• Accessibility audit and fixes
• Performance optimization

Phase 5: Safety & Launch (Weeks 9-10)

• Comprehensive crisis testing
• Security audit
• WCAG compliance verification
• Beta testing with mental health professionals
• Final polish and launch preparation

8. Required Disclaimers

In-App Disclaimer (Always Visible)

┌────────────────────────────────────────────────────────────────┐
│  ⚠️  IMPORTANT NOTICE                                          │
│                                                                │
│  This app is NOT a substitute for professional mental health  │
│  treatment. Aura is an AI companion for wellness support and  │
│  cannot diagnose, treat, or provide medical advice.           │
│                                                                │
│  If you are in crisis, please contact:                        │
│  🆘 Emergency: 911                                             │
│  📞 988 Suicide & Crisis Lifeline: Call or Text 988           │
│  💬 Crisis Text Line: Text HOME to 741741                     │
│                                                                │
│  Always consult qualified healthcare providers for medical    │
│  concerns.                                                     │
└────────────────────────────────────────────────────────────────┘

AI Transparency Statement

"I'm Aura, an AI companion powered by Google's Gemini. I'm here to 
listen, support, and help you practice coping techniques. I learn 
from our conversations to provide more personalized support, but 
I'm not a therapist. Your privacy is protected - conversations 
are processed securely and you control your data."

9. Open Questions

1. AI Avatar Design: What visual style for the Aura character? (Recommendations: soft cloud, gentle creature, abstract shape)

2. GIF Source: Create custom GIFs or use/license existing ones? (Custom recommended for brand consistency)

3. Voice Features: Include text-to-speech for AI responses? (Good for accessibility, adds cost/complexity)

4. Community Features: Include peer support? (Significant moderation requirements)

5. Professional Integration: Direct therapist connections? (Triggers licensing requirements)

6. Monetization: Free forever, freemium, or subscription? (Affects feature gating)