Member-only story

Running Machine Learning Models Directly in the Browser Felt Impossible — Until It Wasn’t

My Experience Building Client-Side AI Systems with WebGPU and TensorFlow.js.

Maximilian Oliver

3 min read2 hours ago

For the longest time, machine learning lived on servers.

• heavy GPUs
• backend APIs
• network latency
• scaling headaches

Every ML feature meant:

frontend → API → model → response

And that worked…

…but it always felt inefficient.

Because sometimes, the best place to run a model is the place where the data already exists:

the browser.

When I started experimenting with WebGPU and TensorFlow.js, it changed how I think about deploying ML systems.

This article is a deep dive into how I build browser-side machine learning systems.

Why Move Machine Learning to the Browser?

At first, it sounds like a bad idea.

Browsers are limited… right?

But here’s what you gain:

• zero latency (no network calls)