We were inspired by the idea that anyone should be able to gage their health and get a quick read on a skin concern without waiting weeks for a dermatologist appointment. We built a custom image classification model using a fine-tuned ResNet18 trained on nearly 500 labeled images of acne, eczema, and healthy skin in Google Colab. The frontend is a React app deployed on Vercel that sends uploaded images through an ngrok tunnel to our Flask API running the trained model in Colab. One of our biggest challenges was connecting the frontend to the backend, dealing with CORS issues, ngrok tunneling, and making sure the base64 image format matched what our model's preprocessing pipeline expected. We also learned how important prompt engineering and response formatting are, our model only outputs a class label and confidence score, so we had to build a mapping layer that translates that into useful descriptions, severity ratings, and care recommendations. Overall, this project taught us how to take a machine learning model from training to deployment as a full-stack application in under 8 hours.
Built With
- flask
- google-colab
- javascript
- ngrok
- python
- pytorch
- react
- vercel
- vite