Le Wagon Bootcamp Project · Nov – Dec 2022
This was my first ever machine learning project, and the one that got me genuinely hooked on AI. It was built during Le Wagon, an intensive data science bootcamp I did in Zurich in late 2022. Le Wagon covers the full stack of practical data science over ten weeks, from Python and SQL through to machine learning, deep learning, and deployment. The final two weeks are dedicated to a team project, and this was ours: AIdentify, a real-time facial emotion recognition system built from scratch in under two weeks with a team of three.
The system detects faces in live webcam video and classifies the expressed emotion across eight categories: neutral, happy, sad, surprised, fearful, angry, disgust, and contempt. We trained on a combined, deduplicated dataset of approximately 40,000 labelled images (FER+ and AffectNet-HQ), applying data augmentation and class weighting to handle label imbalance across categories. A custom CNN baseline reached 71% validation accuracy; fine-tuning a pretrained DenseNet169 via transfer learning, with frozen convolutional weights and a retrained classifier head, pushed that to 81%. The inference pipeline runs OpenCV face detection into a FastAPI prediction endpoint, streamed live through a Flask web app. We presented it at Le Wagon Demo Day in Zurich in December 2022, and it's still the project I point to as the moment I realised I wanted to keep building things like this.