Training a model and running that model on a user's phone without lag are two entirely different jobs. So are reading data from a sensor and turning that data into a meaningful interface. I enjoy closing the distance between the two.
That's why my work has taken me from AI models to embedded systems, from mobile apps to corporate web platforms. On one project I'm writing CUDA-accelerated object detection on a Jetson board; on another I'm building an organisation's entire digital presence from scratch.
The common thread: thinking about a product as a whole rather than a single piece of it — from the database schema to the last pixel of the interface. Some projects I built alone from start to finish, others as part of a team; in both I tried to see the whole picture.