Stroke Shield | Devpost

Inspiration

Strokes are a leading cause of disability and death worldwide, yet early detection can drastically improve outcomes. We were inspired to build Stroke Shield after learning how crucial the first few minutes are in identifying stroke symptoms. Our goal was to create a fast, accessible, and intelligent tool that could help spot early warning signs—potentially saving lives.

What it does

Stroke Shield uses facial recognition and machine learning to scan a person's face for common stroke indicators like facial drooping, asymmetry, or lack of movement. Once a scan is complete, the program flags potential symptoms and alerts the user or caregiver to seek immediate medical attention. The goal is to act as an early warning system and help reduce delays in diagnosis.

How we built it

We built Stroke Shield using JavaScript, React, and HTML for the frontend, styled with TailwindCSS for a clean and responsive UI. For facial landmark detection, we used MediaPipe, which allowed us to track facial movement and symmetry in real time. The backend runs on Express.js, and we integrated Gemini AI for intelligent interpretation and classification of facial features that may indicate stroke symptoms.

Challenges we ran into

One big challenge was accurately interpreting facial landmarks in real time without generating false alarms. Tuning the sensitivity of detection to avoid both under- and over-alerting users took a lot of trial and error. Integrating Gemini AI into our workflow and making it work smoothly with MediaPipe was also a bit tricky, especially when handling edge cases in facial expressions and lighting conditions.

Accomplishments that we're proud of

Even though it’s still in development, we’re proud of getting a working prototype up and running that demonstrates the core idea. The facial tracking component works in real time, and the UI is clean, intuitive, and responsive. Bringing together multiple technologies into one cohesive project was a big milestone for us.

What we learned

We learned a lot about real-time computer vision and how to integrate multiple frameworks like MediaPipe and Gemini AI effectively. We also gained experience in designing accessible user interfaces and making health-related tech more approachable. It was a great lesson in combining AI with web development for real-world impact.

What's next for Stroke Shield

We plan to refine the AI model for better accuracy and train it on more diverse facial data. We also want to add more features—like voice analysis or arm movement detection—to broaden the scope of stroke symptom detection. Long-term, we hope to release Stroke Shield as a mobile app so it can be used on the go, especially in emergency scenarios or remote areas.