Face Recognition

Some projects live in a paper. This one lived in the building.

The goal was a face recognition system good enough to actually use — and that's exactly what it became. I built it with both hands of machine learning at once: classical, feature-based methods and deep learning with convolutional neural networks, letting the two approaches reinforce each other.

The best part wasn't the accuracy on a test set — it was watching it work in the real world. The system recognized faces reliably enough to be put into practice inside our own community, which is a very different bar from "works on the benchmark." Making something people around you actually rely on is its own kind of proof.

Under the hood (for the engineers): a hybrid of conventional feature-based recognition and CNN-based deep learning, tuned for real, practical deployment rather than a leaderboard.

Why it matters: it was one of the moments deep learning stopped being an idea for me and became a tool — something I could point at a real problem and watch it just work. That's a feeling you chase for the rest of a career.

Skills: Deep Learning · Computer Vision · Convolutional Neural Networks · Machine Learning