🧠 Inspiration

The increasing availability of wearable health devices and the urgent need for proactive healthcare inspired us to build a system that goes beyond passive data tracking. We envisioned an intelligent assistant that could provide real-time, personalized health insights—helping users detect potential risks early and take informed action. With the rise of AI and accessible sensors, we saw an opportunity to bridge the gap between raw data and meaningful health advice.

🩺 What it does

AI Health Monitor collects health metrics from wearable devices such as heart rate, blood oxygen, and activity levels and applies machine learning to detect anomalies in real time. It alerts users when irregular patterns arise (e.g., an unusually high heart rate or low oxygen saturation) and provides tailored health recommendations based on the user’s history and trends. All of this is visualized through a clean and intuitive mobile or web interface.

🛠️ How we built it

We followed a modular approach to build the system:

1. Data Simulation & Preprocessing We used simulated data and public datasets like (https://physionet.org/) to model real-world wearable data. Data cleaning and normalization were performed using pandas and numpy.

2. Anomaly Detection Models We implemented anomaly detection using IsolationForest from scikit-learn, and experimented with LSTM networks for sequential prediction using TensorFlow.

3. Web App Interface A lightweight Flask app was developed to simulate real-time updates and visualize data, with plans to expand into mobile using Flutter.

4. Deployment We containerized the app using Docker and prepared for cloud deployment on Azure, with the ML model accessible via a REST API.

from sklearn.ensemble import IsolationForest
model = IsolationForest(contamination=0.1)
df['anomaly'] = model.fit_predict(df[['heart_rate', 'blood_oxygen']])

⚠️ Challenges we ran into

• Data Variability: Health data can be noisy and inconsistent, especially with simulated inputs.
• Model Calibration: Finding the right balance between sensitivity and false positives in anomaly detection required extensive tuning.
• Privacy Considerations: Designing for HIPAA/GDPR compliance added constraints on data handling and storage.
• Real-Time Visualization: Simulating live updates and integrating with front-end dashboards was a performance and architecture challenge.

🏆 Accomplishments that we're proud of

• Built a working pipeline from data ingestion to AI inference to UI alerts.
• Successfully detected anomalies in heart rate and blood oxygen levels.
• Designed an intuitive front-end with a real-time health status view.
• Created a scalable architecture ready for production deployment on cloud platforms.

📚 What we learned

• How to process time-series physiological data for real-world applications.
• The strengths and limitations of using unsupervised ML models like Isolation Forest in health scenarios.
• How to simulate real-time systems and integrate AI into a full-stack solution.
• Best practices for secure, privacy-conscious health tech development.

🚀 What's next for AI Health Monitor

• Integration with Real Wearables: Use Fitbit and Apple Health APIs for live data streaming.
• Advanced Predictive Models: Deploy LSTMs and transformer-based models to forecast health issues.
• User Personalization: Tune recommendations based on user lifestyle and historical patterns.
• Mobile App Launch: Build a cross-platform mobile experience for broader accessibility.
• Clinical Collaboration: Partner with healthcare providers to validate model accuracy and gain medical oversight.

Built With

• api
• apple
• cosmos
• db
• docker
• fitbit
• flask
• github
• healthkit
• javascript
• matplotlib
• numpy
• pandas
• physionet
• plotly
• postgresql
• python
• scikit-learn
• streamlit
• supabase
• tensorflow