Inspiration
One of our teammates lost his grandmother to pneumonia — a disease that was totally treatable, but went undetected. She lived in rural India, where radiologists are rare, and her CT scans were never properly diagnosed. She held onto false hope while the disease silently stole her breath. That pain lit a fire in us. We realized this wasn’t just his story — it’s the story of millions in underserved areas. So we built Ayutra to help patients breathe easier — by catching what the eye might miss, and what time might otherwise take.
What it does
Ayutra is an AI-powered dashboard that detects pneumonia from chest X-rays — fast, accurate, and explainable. One radiologist uploading multiple scans quick and easy, and our model gives a prediction with a confidence score that shows exactly what the AI is seeing. For mass screenings, doctors can upload batches, instantly flag critical cases, and send patients personalized alerts in their own language — no more guesswork, no more silence. Just diagnosis, delivered.
How we built it
We trained a convolutional neural network on a Kaggle chest X-ray dataset with over 8 THOUSAND images for training, using data augmentation and regularization to make sure our model didn’t just memorize, but actually understood. The backend runs on Flask, while the frontend runs on Next.js + React on a typescript server with integrations for email and SMS notifications.
Challenges we ran into
Trust in AI: Doctors won’t use a black box. That’s why we spent time building out explainability with confidence scoring.
Connectivity issues: Rural clinics often have low or unstable internet, so we optimized for low-bandwidth and asynchronous use cases, by optimizing our Artificial Intelligence model to become more lightweight to run on the device itself.
Emotional weight: This wasn’t just a project. Every scan felt like a second chance — and that kind of pressure pushed us to build something real, not just functional.
Accomplishments that we're proud of
Built a CNN model that detects pneumonia with 95% accuracy and strong recall for positive cases.
Designed an app simple enough for village clinics, but powerful enough for real diagnosis.
Turned one radiologist’s workload into a scalable solution — able to screen hundreds with just a few clicks.
What we learned
Accuracy is cool, but explainability builds trust.
Sometimes the best tech is the most invisible.
Healthcare isn’t about numbers. It’s about names, faces, families — and the breath between life and loss.
We learned that impact doesn’t require a hospital. Just a laptop, an idea, and the will to care.
What's next for Ayutra
Add severity detection to prioritize critical patients even faster.
Support WhatsApp notifications, the real lifeline in rural India.
Create offline-first uploads for scan camps with no internet access.
Expand the model to detect TB, COVID-19, and more.
Partner with NGOs and PHCs to test Ayutra in real villages — because code means nothing until it’s saving someone’s breath.
Log in or sign up for Devpost to join the conversation.