Inspiration
Access to basic diagnostic care is a luxury in many rural and underserved areas. People often ignore visible symptoms—rashes, swelling, eye redness—either due to a lack of knowledge or fear of visiting hospitals unnecessarily. We asked ourselves: What if your phone camera could be your first doctor? That question sparked SymptoSnap, a visual symptom analyzer that empowers users to understand their health conditions early and make informed decisions.
What it does
SymptoSnap lets users:
📸 Upload or take a photo of a visible symptom (e.g., skin rash, eye redness).
🧠 Get a real-time ML-based prediction of possible conditions.
🩺 Access reliable information on next steps: self-care vs. see a doctor.
🌐 Optionally connect with a telemedicine provider for follow-up.
🔐 Anonymously store symptom history for future reference or sharing.
How we built it
Frontend: React Native for cross-platform mobile app.
Backend: Flask API serving image classification predictions.
ML Model: Custom-trained CNN on a dataset of skin conditions, eye infections, burns, etc. (e.g., HAM10000, DermaMNIST).
Computer Vision: OpenCV for image preprocessing (resizing, contrast correction).
Deployment: Model hosted on a cloud server using Docker + REST API.
Extras: Integrated Firebase for user auth and Firestore for image & prediction history.
Challenges we ran into
⚠️ Dataset quality: Finding diverse, high-resolution, ethically sourced datasets was tough.
🏷️ Label ambiguity: Many symptoms look similar — achieving reliable model accuracy required thoughtful model design and fine-tuning.
📶 Offline usability: Mobile health apps need to work in low-connectivity areas; we struggled balancing performance vs. model size.
👩⚕️ Medical validation: Ensuring clinical relevance and avoiding misdiagnosis was a critical ethical consideration.
Accomplishments that we're proud of
🚀 Built an end-to-end functional prototype with a trained model in under 3 weeks.
🔍 Achieved 79% classification accuracy on real-world test images.
💡 Developed a modular design that can plug into any healthcare app or EHR system.
🌍 Tested successfully in a simulated rural use case with offline-first support.
What we learned
💬 How to build real-world AI tools that don’t overpromise and prioritize safety.
🧠 Improved skills in image classification, transfer learning (MobileNetV2, EfficientNet).
👨⚕️ Learned to navigate the intersection of healthcare, AI, and ethics.
⚙️ Understood the challenges of scaling health apps, especially in multilingual, low-resource contexts.
What's next for SymptoSnap
🧪 Expand dataset with real doctor-reviewed samples for better accuracy.
🌐 Integrate telemedicine API (e.g., Practo, 1mg) for instant remote help.
🗣️ Add multilingual voice assistant to guide low-literacy users.
🧠 Improve AI with multimodal diagnostics: pair images with symptom questions.
📦 Develop a Lite version with on-device ML for offline-first diagnosis.
Built With
- expo.io
- flask
- html/css
- javascript
- native
- opencv
- python
- react
- tensorflow
- typescript