Writing

My first doorbell model scored an F1 of 0.95 on its test set and still detected doorbells all the time once deployed. Here is how I built it, and why that number was lying.

Data collection is a crucial step for ML project. In the second post of my doorbell detector series I will cover my first attempts and the coldstart problem.

This is the first post of an (hoepfully) upcoming series about my doorbell detection system.

A first post introducing the blog and what to expect here — notes on embedded systems, machine learning, and the things I build.