MindCare - AI-Powered Mental Health Companion

MindCare is a comprehensive, state-of-the-art mental health companion designed to provide emotional support, track user wellbeing, and guide users to professional clinical help when necessary. The platform combines conversational AI with local machine learning models, structured self-assessments, persistent daily journaling, and interactive wellness resources to create a private, supportive, and data-driven emotional support system.

---

Key Features

1. Dynamic AI Companion (Assistant)

• Empathetic Conversations: Underpinned by the Google Gemini API (gemini-2.5-flash), acting as an active-listening, supportive virtual assistant.
• Context-Aware Personalization: Automatically adapts its conversational tone based on the user's latest self-assessment scores and journal sentiment analysis, without ever mentioning the data collection directly.
• Multilingual Script Support: Natively converses in:
  • English
  • Roman Urdu (Urdu words in English letters)
  • Urdu script It dynamically matches the user's script choice and enforces strict script alignment rules.
• Strict Boundary Framework: Safe-guarded with strict boundaries. It prevents self-diagnosis, avoids therapy replacement claims, and intelligently recommends nearby clinics only during high-severity scores or crisis states.

2. Structured Self-Assessment

• Mental Wellbeing Testing: Uses standard assessment scales to evaluate user levels of Anxiety, Depression, and Stress.
• Historical Tracking: Saves all questionnaire attempts in a persistent database, letting users visualize their emotional progress over time on an analytical dashboard.

3. Daily Journaling & Weekly Analytics

• Persistent Daily Logs: Allows users to document their daily thoughts. Subsequent inputs on the same day append to the existing log to maintain daily coherence.
• Local Hugging Face NLP Classifier: Integrates a locally hosted transformer classification pipeline that parses journal entries into four states: Normal, Anxiety, Stress, and Depression.
• Polarity Snapshots: Computes numerical polarity scores dynamically.
• Weekly Analytical Reports: Tracks weekly sentiment trends (improving, stable, worsening) and maps out dominant emotional state distributions.

4. Interactive Meditation & Guided Wellness

• Breathing Exercises: Interactive, visual guidance pages for:
  • Deep Breathing
  • Breath Focus
  • Square Breathing (calming box method)
• Guided Meditations: Mindful tracks including Body Scan, Mindful Awareness, Loving Kindness, Progressive Relaxation, and Calm Place Visualization.
• Grounding Techniques: Interactive guides for the 5-4-3-2-1 Technique and Progressive Muscle Relaxation.
• Relaxing Audios: High-fidelity nature ambient audios (Ocean Waves, Rain Forest, Mountain Stream, Forest Birds, Rain & Thunder, Sea Waves) built with a custom audio control system.

5. Clinic Locator

• Geolocated Search: Helps users locate verified clinics nearby based on coordinates.
• Distance Calculation: Employs the Haversine formula to calculate precise distances between the user's coordinates and registered clinic locations.

6. Profile Customization & Security

• Secure Authentication: Implements JWT-based access and refresh tokens.
• Security Updates: Supports secure password change operations and code-verified password reset via SMTP email routing.
• Personalized UI: Custom dark/light mode provider and email preference toggling.

---

Technology Stack

graph TD
    subgraph Frontend [React Frontend client]
        A[Vite + React & TypeScript] --> B[Material UI & Tailwind CSS]
        A --> C[Recharts Analytics]
        A --> D[React Router DOM]
    end

    subgraph Backend [FastAPI Backend Server]
        E[FastAPI Web Framework] --> F[SQLAlchemy ORM]
        E --> G[Hugging Face Transformer Pipeline]
        E --> H[Google Gemini SDK]
        F --> I[(PostgreSQL DB)]
    end

    A -- REST API Calls --> E

Loading

Frontend

• Framework: Vite + React (TypeScript)
• Styles: Tailwind CSS + Material UI (MUI) for a premium, responsive experience
• Charts: Recharts for historical score graphs
• Routing: React Router v7
• Icons: Lucide React

Backend

• Core Framework: FastAPI (Python 3.10+)
• Database ORM: SQLAlchemy with PostgreSQL (Neon DB compatible)
• Database Migrations: Alembic
• Email Server: FastAPI-Mail (SMTP wrapper)
• Authentication: PyJWT, Cryptography (Argon2 hashing)

Machine Learning / AI

• Generative AI: Google Gemini Developer API (google-genai SDK)
• Local Classification: Hugging Face transformers pipeline loaded with a local text classification model (safetensors format) under /backend/model.

---

Key Directory Structure

MindCare/
├── backend/
│   ├── alembic/              # Database migration history
│   ├── app/
│   │   ├── dependency/       # Auth dependency middleware
│   │   ├── models/           # SQLAlchemy Database Models
│   │   ├── routes/           # FastAPI API Endpoints
│   │   ├── schemas/          # Pydantic Schemas (validation)
│   │   ├── services/         # Business logic (Gemini, Inference, Assessment, etc.)
│   │   ├── db.py             # DB connection initializer
│   │   ├── main.py           # FastAPI main app entry point
│   │   └── settings.py       # Pydantic Settings & ENV parser
│   ├── model/                # Local Hugging Face Model files (config, model.safetensors, tokenizer)
│   ├── scheduler.py          # Cron/Manual script for weekly report generation
│   ├── requirements.txt      # Python dependencies
│   └── pytest.ini            # Test configurations
└── frontend/
    ├── public/               # Static assets (audios)
    ├── src/
    │   ├── app/
    │   │   ├── components/   # Pages & shared UI components
    │   │   ├── context/      # AuthContext & state providers
    │   │   ├── App.tsx       # Root Component
    │   │   └── AppRoutes.tsx # Client Route configurations
    │   ├── styles/           # Global styles and custom CSS variables
    │   └── main.tsx          # Frontend entry point
    ├── package.json          # Node dependencies & scripts
    └── vite.config.ts        # Vite build configurations

---

Getting Started

Prerequisites

• Python 3.10+
• Node.js (v18+)
• PostgreSQL database instance (local or hosted, e.g., Neon DB)

---

Backend Setup

1. Navigate to the backend directory:

   cd backend

2. Create and activate a virtual environment:

   python -m venv venv
   # On Windows (PowerShell):
   .\venv\Scripts\Activate
   # On macOS/Linux:
   source venv/bin/activate

3. Install python packages:

   pip install -r requirements.txt

4. Create a .env file in the root of the backend/ directory following this format:

   DATABASE_URL="postgresql://<username>:<password>@localhost:5432/MindCare"
   ACCESS_TOKEN_SECRET_KEY="your-jwt-access-secret"
   REFRESH_TOKEN_SECRET_KEY="your-jwt-refresh-secret"
   ALGORITHM="HS256"
   ACCESS_TOKEN_EXPIRE_TIME=20
   REFRESH_TOKEN_EXPIRE_TIME=10
   
   MAIL_USERNAME="your-email@gmail.com"
   MAIL_PASSWORD="your-email-app-password"
   MAIL_FROM="your-email@gmail.com"
   MAIL_SERVER="smtp.gmail.com"
   MAIL_PORT=587
   MAIL_STARTTLS=True
   MAIL_SSL_TLS=False
   
   CORS_ORIGIN="http://localhost:5173"
   GEMINI_API_KEY="your-google-gemini-api-key"
   GEMINI_MODEL="gemini-2.5-flash"

5. Run database migrations:

   alembic upgrade head

6. Start the FastAPI server:

   uvicorn app.main:app --reload

   The backend will launch at http://localhost:8000 with Swagger docs available at http://localhost:8000/docs.

---

Frontend Setup

1. Navigate to the frontend directory:

   cd ../frontend

2. Install Node dependencies:

   npm install

3. Launch the Vite development server:

   npm run dev

   The frontend will run at http://localhost:5173.

---

Weekly Analytics Scheduler

To process user daily journals and compile weekly mental health reports, a background scheduler is provided. This analyzes diary logs created during the current calendar week (since Monday 00:00:00).

• Process all active users for the current week:

  python scheduler.py

• Process a specific user by ID manually:

  python scheduler.py --user_id <USER_ID>

  (e.g., python scheduler.py --user_id 1)

---

Security & Privacy Notice

MindCare respects user privacy. Journal entries and self-assessment history are stored in a private database. AI assistant context analysis is computed during runtime session requests and is not permanently trained on by external large language models.