🧠 StressTracker AI

Your digital wellbeing, monitored in real-time.

StressTracker AI is a privacy-focused, local-first application designed to monitor digital biomarkers—such as mouse movements and keystroke dynamics—to detect stress and cognitive load in real-time. By leveraging local AI, it provides personalized insights without compromising your data privacy.

---

✨ Features

• 🖱️ Kinematic Analysis: Detects Sympathetic Nervous System (SNS) activation by tracking mouse jitter, path efficiency, and velocity.
• ⌨️ Keystroke Dynamics: Assessments cognitive load by analyzing flight time variance (ISO 9241-11).
• 🤖 Dr. AI Assistant: A Llama-3.2 powered clinical psychiatrist persona offering actionable, text-based feedback.
• 📊 Real-time Dashboard: An immersive "Deep Space" themed UI built with Streamlit for monitoring your status.
• 🔒 Privacy First: Completely local processing. No raw input data ever leaves your device.

⚡ Performance & Benchmarks

The core feature extraction pipeline is optimized for real-time processing with minimal overhead.

Benchmark Results:

• Throughput: ~923,000 events/second
• Processing Time: 100,000 events processed in ~0.11 seconds
• Efficiency Calculation: Combined path efficiency algorithms execute in < 0.8s for large datasets.

Tested on Silicon-based Architecture.

---

🚀 Getting Started

Prerequisites

• Python 3.9+
• Ollama: Required for the AI agent. The application will automatically pull the llama3.2 model if needed.

Installation & Run

We provide a startup script that handles virtual environment creation, dependencies, and launching the application.

1. Clone the repository:

   git clone https://github.com/alwaysvivek/stress-tracker.git
   cd stresstracker

2. Launch the application:

   python start.py

The application will automatically open in your default web browser.

---

📊 Dashboard Overview

Main Interface

Analysis & Insights

Real-time Metrics	Dr. AI Consultation

📘 Technical details: For a deep dive into the mathematical models and signal processing, please refer to the Methodology & Signal Processing Documentation.

---

🏗️ Architecture

The system operates as a monolithic local application to ensure system-wide tracking and privacy.

graph TD
    User[User Input] -->|Hooks| Listener[pynput Listener]
    Listener -->|Events| Tracker[core.tracker.BackgroundTracker]
    Tracker -->|Raw Data| Memory[In-Memory Buffer]
    
    Memory -->|Micro-Batch| Extractor[core.features.FeatureExtractor]
    Extractor -->|Features| Analysis[core.analysis.submit_session]
    
    Analysis -->|Z-Scores| Agent[core.agent.StressManagementAgent]
    Agent -->|LLM Inference| Ollama["Ollama (Llama 3.2)"]
    
    Analysis -->|Results| UI[Streamlit Frontend]
    Agent -->|Advice| UI
    UI <-->|Chat| Agent

Loading

Why "Local Only"?

Important

System-Wide Tracking: To accurately analyze work stress, the application must track inputs across all applications (Excel, Slack, IDEs, etc.). Web browsers restrict this access for security reasons. Thus, StressTracker AI runs locally to utilize system-level hooks via pynput while keeping your data secure on your machine.

---

🔧 Troubleshooting

Common Issues

1. "Ollama not found"

• Cause: Ollama is not installed or running.
• Solution: Install from ollama.com and verify with ollama list in your terminal.

2. "Input monitoring permission denied" (macOS)

• Cause: macOS security restrictions blocking keystroke monitoring.
• Solution:
  1. Navigate to System Settings > Privacy & Security > Input Monitoring.
  2. Remove PyCharm (or your terminal/IDE) using the minus (-) button.
  3. Restart the application and click Grant Permission when prompted.
  4. Restart your IDE/Terminal completely.
• Advanced Fix: Run tccutil reset Accessibility in the terminal and potential restart your computer.

3. "ModuleNotFoundError"

• Cause: Dependencies not installed in the current environment.
• Solution:

  source .venv/bin/activate
  pip install -r requirements.txt

---

Built with ❤️ for Digital Health