What if building for constraints isn't about making a weaker system, but designing a better one?

In this talk, I share lessons from a personal project where a 3 MB payload was reduced to just 60 KB by rethinking how the problem was represented and where computation should happen.

Rather than relying on heavier models, larger servers, or sending raw visual data to the cloud, the project explores MediaPipe, WebAssembly (WASM), and on-device processing to build a system that is smaller, faster, more private, and better suited to environments with limited resources.

The talk uses this experience to explore a broader approach to engineering: design intentionally around the constraints of your users, devices, data, and environment. The key decisions made were around what information really needed to be kept/discarded , how much data we needed to transmit and how we could reduce user waiting time.

The goal isn't simply to make ML smaller. It's to ask a better question: What is the minimum system we can design that still solves the problem well?