Problem
Therapy and wellness apps mostly rely on what people say or type about their feelings. Our team wanted a device that could pick up emotional signals directly, without asking. Infrared cameras are useful here because they capture heat patterns on the face, not only how it looks.
What I built
The idea was inspired by a research paper, NeckFace (Chen et al., 2021), on reconstructing facial expressions from infrared images of the chin and jaw. We wanted to turn that into a small device.
As a team we recorded first infrared videos and tried out simple approaches on them. I built the machine learning baseline: a ResNet50 model that predicts one of seven emotions from face images, trained on the public FER2013 dataset. That script is in the public repository. I also worked on the 3D-printed hardware prototypes and helped design the user interface in Figma.
My part
We were six people with different backgrounds. I worked on the machine learning side and supported the hardware and the UI design. The business side and the overall product were a team effort.
Result
The project won Best Startup Idea in Berlin and received the TU Scholarship.
What I’d do next
Collect a larger infrared dataset recorded with our own hardware and test the model on people it has never seen.