DEV Community: Leman Caliskan

Restoring the Rhythm of Rest: Introducing SleepMetrics

Leman Caliskan — Mon, 06 Apr 2026 14:30:35 +0000

Honestly, I never planned to build something like SleepMetrics. Not at first.

I developed CardioMetrics-Core what is an AI-powered desktop application that analyzes clinical data to evaluate cardiovascular risk levels and provide real-time health insights using machine learning models.

And I enjoyed developing the CardioMetrics project. And I thought, could I create a similar application related to health? And that’s how Sleepmetrics was created.

✨ My Approach

I wanted SleepMetrics to be more than just a spreadsheet of numbers. The tool needed to be simple, functional, and professional.

CustomTkinter ended up being a surprisingly good fit once again. I maintained the light/dark modes to ensure the interface felt comfortable to use especially important for an app you might check at night.

Real-time Visualization: I integrated live gauges that react to your data. When you adjust stress levels or physical activity, the app provides immediate visual feedback.
Personalized Sleep Insights: The app doesn't just show numbers; it provides specific guidance based on sleep duration, activity levels, and stress to help you optimize your rest.

🩺 Smart Health Recommendations

Based on the lifestyle and clinical data entered, SleepMetrics generates a dynamic dashboard of personalized suggestions. These aren't generic tips; they are reactive nudges triggered by your specific metrics:

🧠 Disorder & Quality Alerts

SleepMetrics transforms your data into three different health statuses based on the machine learning model’s findings:

Sleep Apnea Risk Detected 🚨: If the model identifies patterns consistent with Apnea, the app triggers a high-priority alert (Red) recommending a consultation with an ENT specialist.
Insomnia Symptoms Detected ⚠️: When insomnia markers appear, the app provides practical tips, such as cutting out blue light sources 2 hours before bed to help your brain naturally produce melatonin.
Healthy / Asymptomatic ✅: If no risks are detected, the app encourages you to keep up the good work, confirming your data aligns with a healthy sleep profile.

🔍 Parameter Specific Health Recommendations

Stress Management 🧘: If stress levels exceed 7/10, the app suggests pre-sleep breathing exercises to lower cortisol levels.
Sleep Duration ⏰: Logging less than 6 hours triggers a warning about the lack of physical repair and encourages a 7+ hour target.
Activity & Movement 👣: Low step counts (< 5,000) or low activity trigger a reminder that reaching 7,500 steps can shorten the time it takes to fall asleep by 15%.
Heart Rate & BP ❤️: High readings prompt advice to reduce caffeine intake and monitor vitals more closely.

Note: They’re not the a substitute for professional medical advice, diagnosis, or treatment. It just a simple advices and friendly nudges to support your heart health.

🛠 The Tech Stack & Model

For the backend, I used these datasets Sleep Health and Lifestyle and Sleep Cycle and Productivity from Kaggle.

I built a dual-engine system:

Quality Prediction: A RandomForestRegressor (150 estimators) handles the sleep quality score with a 99.08% $R^2$ accuracy.
Diagnosis: A GradientBoostingClassifier predicts potential sleep disorders with 88% accuracy.

The model provides a signal: "maybe check this out."

🌍 Why Two Languages? (EN/TR)

As I’m Turkish, I developed the app in Turkish first. But health is universal. I restructured the project to have clean v_EN and v_TR folders so language would never be a barrier to access.

🖥️ Choose Your Experience: (Web, Desktop, or Code)

People have different preferences. Some want instant access, some prefer a standalone desktop app, some just want the code.
So I made three options:

Streamlit Web App: quick access → SleepMetrics Streamlit App
Desktop App: A pre-built .exe file (no Python environment required). Just download and run.
Source Code: Cleanly structured folders with clear EN/TR separation for the curious.

🚀 Explore the Project

lemancaliskan / SleepMetrics

Sleep Health & Lifestyle Analysis Tool: An AI-powered desktop application that analyzes biometric data to predict sleep quality and detect potential sleep disorders using machine learning models.

🌙 SleepMetrics - Sleep Health & Lifestyle Analysis Tool

SleepMetrics is a modern desktop application designed to analyze sleep quality and predict potential sleep disorders using advanced machine learning algorithms. By processing lifestyle and clinical data through a sleek, high-DPI interface, it provides users with data-driven insights into their sleep health.

📺 Demo

🎨 Visual Experience

The application features a dedicated toggle for seamless switching between light and dark modes.
Features integrated real-time gauges for the live visualization of sleep quality scores.

🔍 Desktop Application (EN/TR)

Optimized for a 980x666 centered window layout, this standalone application delivers a precision-focused, localized experience through a theme-aware CustomTkinter UI designed for both global and local users.

TURKISH UI:

ENGLISH UI:

🌐 Web Application (Streamlit):

A responsive and lightweight web version for instant access from any device.

✨ Features

Dual Language Support: Optimized interfaces for both English (EN) and Turkish (TR).
Modern GUI: A…

View on GitHub

Streamlit Web App: Try it here
Desktop App: Pre-built .exe available in GitHub Releases.

🤝 Contributing

If you have ideas on how to improve the ML model or want to suggest new features for the UI, feel free to open an issue or a Pull Request on GitHub. Let’s build better health tools together!

Medical Disclaimer: This software is for informational purposes only. The results provided do not constitute a formal medical diagnosis. Always consult with a professional healthcare provider before making any medical decisions.

Bridging the Gap Between Data and Diagnosis: CardioMetrics Core

Leman Caliskan — Mon, 06 Apr 2026 11:48:55 +0000

Honestly, I never planned to build something like CardioMetrics Core. Not at first.

It all started when I went to the hospital’s cardiology clinic. The waiting room was packed. People with high blood pressure, heart issues, so many patients. It was overwhelming.

And I kept thinking: if all these people need monitoring, why isn’t there a simple, accessible way to keep track of heart health at home? So I decided to try creating something myself.

✨ My Approach

I wanted the tool to be simple and functional. Fast, responsive, and easy to interact with—nothing fancy, but still professional.

CustomTkinter ended up being a surprisingly good fit. I added light/dark modes, and the interface started to feel comfortable to use.

One of the key things I focused on was real-time feedback. When users enter values, the app compares parameters like:

Systolic Blood Pressure
Resting Heart Rate
Cholesterol
Fasting Blood Sugar

If a measurement is outside the reference range, the progress bar changes color (e.g., Red for high risk, Green for normal).

🩺 Smart Health Recommendations

Based on the data entered and the calculated risk score, CardioMetrics Core provides personalized suggestions. These aren't generic tips; they respond directly to your measurements:

📊 Score-Based Alerts

Critical Risk (Score > 70) 🚨: High cardiovascular risk detected. It’s strongly recommended that you consult a cardiologist immediately and start keeping a regular record of your vitals.
Moderate Risk (Score 31–70) 👟: Your heart health shows some areas to watch. A daily 30-minute brisk walk and a Mediterranean-style diet (reducing salt and saturated fats) can make a huge difference.
Low Risk (Score ≤ 30) ✅: Your results look good! Maintain your health with regular sleep, hydration, and annual check-ups.

🔍 Parameter Specific Alerts

Blood Pressure: High (above 140 mmHg) triggers salt intake warnings, while Low (below 90 mmHg) suggests better hydration.
Blood Sugar: High readings (above 120 mg/dL) suggest reducing carbohydrate intake and discussing lifestyle adjustments with a doctor.
Cholesterol: High levels (above 240 mg/dL) prompt advice to avoid high-fat foods, while borderline levels suggest dietary shifts.

Note: They’re not the a substitute for professional medical advice, diagnosis, or treatment. It just a simple advices and friendly nudges to support your heart health.

🛠 The Tech Stack & Model

For the backend, I used the Heart Disease Dataset.

Core: Python, scikit-learn
Models: Logistic Regression, Random Forest
UI: CustomTkinter (Desktop) & Streamlit (Web)

The model sits at around %68.29 accuracy. It’s not a medical diagnosis tool; it’s a signal: “maybe check this out.”