Running ML in the Browser: How I Built MushFinder with ONNX.js, Vite, and PWA

In my first six months of exploring JavaScript, one of the most surprising discoveries was that machine learning doesn’t always need a server. With ONNX.js (onnxruntime-web), it’s possible to run pre-trained models directly in the browser.

In this talk, I wanna showcase MushFinder, a mushroom recognition app I built and deployed as a PWA with Vite. I will explain the basics of ONNX, share how I integrated the model into a front-end app, and discuss what I learned about performance, offline use, and the limitations of running ML at the edge.