MentraAI

MentraAI is an innovative, emotionally intelligent chatbot designed to provide compassionate mental health support and resources to individuals experiencing distress. By leveraging advanced language models and emotion analysis, MentraAI delivers empathetic and context-aware conversations that help users feel genuinely heard, understood, and supported.

In addition to its core mental health support capabilities, MentraAI also integrates a visual diagnostic feature. This allows users to upload images of visible ailments—such as pimples, rashes, or skin conditions like measles—for preliminary analysis. Utilizing cutting-edge image recognition technology, MentraAI can provide a diagnostic suggestion with up to 79% accuracy, helping users better understand their symptoms and make informed decisions about seeking professional care.

---

Tech Stack

Frontend Frameworks

• React (with Vite)
• Tailwind CSS
• Next.js

Backend

• Express.js
• MongoDB (used for user authentication and storage)

---

Features

• Empathetic Chatbot: Engages in supportive, emotionally-intelligent conversations.
• Emotion Analysis: Detects and responds to user emotions to provide tailored responses.
• Dynamic Conversation Flow: Adjusts conversation strategy based on user mood and conversation stage (e.g., greeting, support, crisis).
• Configurable Behavior: Users can tune empathy level, response style, and emotion sensitivity.
• Supports Multimodal Inputs: Can process text, and (in advanced configs) voice and visual content.
• Document Ingestion: Supports ingestion of various document types for resource enrichment.
• Visual Diagnostics: Users can upload images for preliminary analysis of visible ailments.

---

Directory Structure

• src/mentraai/ - Core implementation of MentraAI, including chatbot logic and emotion analysis.
• src/mentraai.egg-info/ - Python package metadata.
• aiqtoolkit-opensource-ui/ - (Optional) UI tools or components.
• configs/ - Configuration files.
• data/ - Example or production data.
• mentra_landpage/ - Landing page resources.
• pyproject.toml - Python project configuration.

---

How It Works

• Emotion Detection: User input is analyzed for primary and secondary emotions (anger, fear, sadness, etc.) along with emotional intensity.
• Context Tracking: Maintains conversation context, user mood history, and interaction counts to personalize responses.
• Adaptive Response: EmpathicChatbot dynamically generates responses based on current emotional and conversational context.
• Crisis Handling: Recognizes high-intensity negative emotions and escalates to crisis support if needed.
• Visual Diagnostics: Users can upload images for AI-powered preliminary health analysis.

---

Advanced Configuration & Usage

Developer-Oriented Configuration

The core chatbot is highly customizable via the MentraaiFunctionConfig class:

• Model Selection: You can specify different models for empathy reasoning, vision analysis, voice input/output, and conversation flow.
  • empathy_reasoning_model: Primary model for empathetic reasoning
  • multi_modal_vision_model: Model for visual content analysis
  • user_voice_model: Model for voice input processing
  • agent_voice_output_model: Model for voice response generation
  • topic_control_model: Model for conversation flow management
• Behavioral Parameters:
  • max_history: Maximum conversation history tokens (default: 100; range: 10–500)
  • emotion_sensitivity: Emotion detection sensitivity (float 0.1–1.0, default: 0.7)
  • empathy_level: "low", "medium", or "high" (default: "high")
  • response_style: "supportive", "professional", "casual", or "therapeutic" (default: "supportive")
  • chunk_size: Document chunk size for analysis (default: 1024; range: 256–4096)
  • auto_analyze: Enable automatic content analysis (default: True)
  • embedder_name: Reference to the embedder model for document ingestion
  • ingest_glob: Glob pattern for ingesting document types (default: **/*.{txt,md,pdf})
  • description: Description of the assistant (default: "Empathic AI Assistant")
• Validation: Custom validators ensure only allowed values for empathy level and response style.

Dataset Integration

• Dataset Management: The backend integrates with health and mental health symptom datasets for improved empathy and diagnostic accuracy.
• Dataset-driven Responses: Responses can be tailored using both conversational context and external knowledge sources.

Multi-Modal Support

• Text, Voice, and Image: The architecture is ready for text, voice, and image input/output, with models selected per capability.
• Document Ingestion: Customizable ingestion patterns for adding resources (see ingest_glob).

Component-Oriented API

• Entry Points: Components are registered via the AIQ Toolkit framework (see src/mentraai.egg-info/entry_points.txt).

---

Setup

Prerequisites

• Python 3.8+
• NVIDIA API key for LLM endpoints (set as NVIDIA_API_KEY in your environment)
• Node.js (for frontend and backend components)
• MongoDB instance (local or cloud)
• Required Python packages (see pyproject.toml)
• (Optional) dotenv for environment variable management

Installation

1. Clone the repository:

   git clone https://github.com/Oloyede-Michael/MentraAI.git
   cd MentraAI

2. Install Python dependencies:

   pip install -r requirements.txt

   Or, if using Poetry or PDM:

   poetry install
   # or
   pdm install

3. Install Frontend and Backend dependencies:

   # For React/Vite/Next.js frontend
   cd aiqtoolkit-opensource-ui
   npm install
   
   # For Express backend
   cd ../backend
   npm install

4. Set environment variables:

   • Create a .env file or export required variables:

     NVIDIA_API_KEY=your_nvidia_api_key
     MONGODB_URI=your_mongodb_connection_string
     

---

Running the Application

• Python Backend (Empathic AI):

  • The core logic is in src/mentraai/mentraai_function.py. The chatbot can be integrated into a service, UI, or run as a standalone process (with custom CLI/web integration).

  Example (Python):

  from mentraai.mentraai_function import EmpathicChatbot, MentraaiFunctionConfig, Builder
  
  config = MentraaiFunctionConfig(
      # configure as needed
  )
  builder = Builder()
  bot = EmpathicChatbot(config, builder)
  # Use bot.process_conversation(...) for chat

• Frontend (React/Vite/Tailwind or Next.js):

  • From the frontend directory:

    npm run dev

  • (See individual frontend README.md for further instructions.)

• Node/Express Backend:

  • From the backend directory:

    node index.js

    or

    npm run start

---

Configuration

The chatbot is highly configurable. Key options include:

• empathy_level: "low", "medium", or "high"
• response_style: "supportive", "professional", "casual", "therapeutic"
• emotion_sensitivity: Float (0.1–1.0)
• max_history: Number of conversation history tokens to keep
• chunk_size: Document chunk size for analysis (default: 1024)
• auto_analyze: Enable/disable automatic content analysis

Edit these settings in your initialization or configuration files.

---

Security Policy

Security is a priority for this project. If you discover a vulnerability or exposed secret:

• Do not disclose vulnerabilities or secrets publicly.
• Contact the repository maintainer(s) privately with details.
• Wait for a response and cooperate as needed for remediation.
• Patches will be documented in release notes, and reporters can be credited if desired.

For full details, see aiqtoolkit-opensource-ui/SECURITY.md.

---

Contributing

1. Fork the repo and clone your fork.
2. Create a new branch: git checkout -b my-feature
3. Make your changes and commit: git commit -am 'Add new feature'
4. Push and open a Pull Request.

See aiqtoolkit-opensource-ui/CONTRIBUTING.md for detailed contributing guidelines, code style, and PR requirements.

---

License

This project currently does not specify a license. Please contact the repository owner for intended usage or contribution terms.

---

Acknowledgments

MentraAI is built with:

• Langchain
• NVIDIA AI endpoints
• Python ecosystem (Pydantic, dotenv, etc.)
• React, Vite, Tailwind CSS, Next.js
• Express.js, MongoDB

---

For questions, issues, or feature requests, please open an issue on GitHub.